WO2002069265A1 - Procede et dispositif de reperage d'un objet par sa forme, sa dimension et/ou son orientation - Google Patents
Procede et dispositif de reperage d'un objet par sa forme, sa dimension et/ou son orientation Download PDFInfo
- Publication number
- WO2002069265A1 WO2002069265A1 PCT/EP2002/001899 EP0201899W WO02069265A1 WO 2002069265 A1 WO2002069265 A1 WO 2002069265A1 EP 0201899 W EP0201899 W EP 0201899W WO 02069265 A1 WO02069265 A1 WO 02069265A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- signal
- space
- locating
- interest
- data
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
Definitions
- the present invention relates to a method and a device for locating an object by its shape, its size and / or its orientation.
- Such methods and devices are already known and, in particular, some of them have proposed the statistical analysis of the points or pixels of a digital video signal coming from an observation system, for the production of effective devices capable of to operate in real time.
- US Pat. No. 5,795,306 discloses a body condition detection device suitable for analyzing images of a face. The pupil is extracted from the images and its characteristics make it possible to judge the body condition.
- the object of the invention is to propose a method of locating an object by its shape, its size and / or its orientation using a device consisting of a set of units for calculating histograms.
- the object is proposed to hierarchically decompose the object to be identified according to its properties. For example, we can first determine the general outline of a moving object in relation to a relatively stable background, then look inside this outline for characteristic elements by their hue, color, relative position ... This approach allows the rapid development of multiple applications involving the identification of an object. These applications can be developed, either from an earlier formalization having identified the significant characteristics of the object, or, thanks to a learning function by examining a scene in which the object in question is present. , the device allowing itself to extract parameters characteristic of the object.
- the invention relates to a method of locating a shape in a space represented by pixels which together form a multidimensional space i, j, evolving over time, and represented at a succession of moments T, said data each associated to a time parameter A, B, ... being in the form of DATA (A), DATA (B), ... digital signals consisting of a sequence A yt , B yt , ... of binary numbers of n bits associated with synchronization signals making it possible to define the moment T of the space and the position i, j in this space, at which the signals A y -t, B y t, ... received at a given instant.
- an area of interest in space is identified as a function of a statistical criterion applied to a time parameter, b) the main area thus identified is inhibited, c) steps a) are repeated b) so as to identify other areas of interest within an area of uninhibited space, d) the process is stopped when a remaining area, uninhibited, of space no longer produces of area of interest corresponding to the statistical criterion, e) a counter for each area of interest thus identified is incremented by consecutive valid frame, the barycenter of its point cloud, f) we recover for each area of interest thus identified , the barycenter of its point cloud.
- a counter is associated with each area of interest, and the counter value is incremented by one. unity at each of the successive frames for which said area of interest is identified, the value of this counter being forced to zero at the first frame for which said area is no longer identified,
- the area of interest is validated for a value of its associated counter greater than 1,
- the validated area is identified by its barycenter, the orientation of its projection axes and the dimensions of the associated frame,
- the main axes of the frame and its dimension are the position, the orientation and the dimension of the object perceived respectively,
- step of claim 2 is repeated, a secondary zone playing the role of the zone of interest, which leads to the identification of tertiary zones,
- the time parameter is a luminance level
- - the time parameter is a color
- the time parameter is the spatial resolution
- the time parameter is the depth of field
- the identified area is defined with respect to a reference chosen from several reference points with different orientations, • - the relative positions of the barycenters of the point clouds of the identified areas of interest are used to control the shape of the object perceived,
- the invention also relates to a device for locating a shape in a space represented by pixels which together form a
- DATA (B) ... numerical in the form of a sequence A yt , Byt, ... of binary numbers of n bits associated with synchronization signals making it possible to define the moment T of space and the position i, j in this space, at which the signals A, Byt, ... received at a given time.
- the device for locating a shape comprises two sets of stogram calculation units receiving the signals and each producing a classification value, the first subset receiving a signal carrying a first time parameter and the second subset receiving two spatial signals, the classification value of the first subset validating a group of points in space processed by the second subset, the number of said points being ni, the classification value of the second subset validating the parameter values processed by the first subset, the two subsets jointly producing a binary signal ZA representing an area of interest and a signal P representing the value of the time parameter in this area.
- It comprises a third subset receiving a signal carrying a second time parameter, this third subset having an operation analogous to the first and replacing it when it validates points in space whose number is n 2 , ⁇ . 2 being greater than ni.
- H includes several subsets receiving spatial signals allowing successive validation of several groups of points in space.
- It includes a set of histogram calculation units controlled by API software and linked together by a data bus and by a feedback bus.
- FIG. 1 is the representation of a histogram calculation unit of the prior art implemented in the invention: - Figure 2 is the representation of the association of two calculation units fulfilling a spatial composite function according to the invention;
- - Figure 3 is the synthetic representation of the device of Figure 2;
- - Figure 4 is a developed representation of the association of a set of histogram calculation units used for the detection and location of an object in black and white;
- FIG. 5 is a developed representation of the association of a set of histogram calculation units used for detecting and locating an object in color
- FIG. 6 is a synthetic representation of the device of Figure 5;
- - Figure 7 shows the combination of the device of Figure 3 with a histogram calculation unit applied to the movement;
- - Figure 8 shows the location of an area of interest;
- Figure 9 is the representation of the histograms, respectively of the movement 9A, of X 9B and of Y 9C for the area of interest of the figure
- - Figure 10 is a flowchart for locating an area of interest
- - Figure 11 shows a device for locating three areas of interest from a main area
- FIG. 12 illustrates the operation of the device of Figure 11 in the form of a graph
- FIG. 13 is a schematic representation of the analysis of the shape formed by a face
- the reference 1 designates a block histogram processing unit as a whole.
- the DATA parameter (A) mentioned by the histogram calculation unit 1 is mentioned by the function 3 (fog).
- This histogram processing unit 1 includes an analysis output register reg 4 containing values representing the statistical distribution of the parameter DATA (A).
- the histogram processing unit 1 supplies the bus 110 with an output value S, it is controlled by the API 5 which determines both the parameter
- This histogram processing unit supplies at output, on the bus 110 a classification value and receives as input, from this same bus 110, the classification values coming from its counterparts with which it is associated.
- the classification value is the result of the application of the function (fog) following values Ay t constituting DATA (A).
- the register reg 4 is the result of different statistical processing of the Ai Jt sequence.
- a preliminary step, facilitating the processing of the signal, whatever the object or its characteristics to be identified, is advantageously provided for and consists in the adaptation of the luminance level and as well as in an adaptation of level and dynamics for location of d 'a moving object.
- the bus 110 provides the histogram calculation units 10 and 11 with the information to be processed and an AND operator 12 combines the output signals • S px and S py . The output of this operator is directed to bus 110.
- Figure 3 is a synthetic representation of the same function. It is easily understood that the set of two analysis units described above and represented in FIG. 2 fulfills the same function as the block of the same nature, and also a histogram calculation unit 6.
- FIG. 4 it is possible to associate a larger number of histogram calculation units 16 to 19 each processing a parameter.
- the luminance parameter is processed by the histogram calculation unit 16
- the saturation signal is processed by the histogram calculation unit 17
- the spatial parameter X is processed by the histogram calculation unit 18
- the spatial parameter Y is processed by the histogram calculation unit 19.
- FIGS. 5 and 6 also show a grouping of units for calculating histograms enabling the color of a selected object to be identified.
- the histogram calculation units respectively 20. for the hue, 21 for the saturation, 22 for the luminance, provide 5 output signals combined by the operator AND 23. This operator provides as output a combined retroannotation signal representative of the color analyzed.
- This complex assembly of FIG. 6 fulfills a function analogous to the histogram calculation unit 24 shown in FIG. 5.
- FIG. 7 two histogram calculation units 100 and 101 are shown, respectively processing the movement parameter. and the spatial parameters X and Y.
- histogram calculation units 100 and 101 are associated by the line 103 corresponding to the bus 110 of the figures described above. 15
- the association of histogram calculation units respectively processing a time parameter, here the movement 'MVT presented by way of example, and spatial parameters such as X and Y, allows the location of an area in the space (i, j). in which a point is represented by the parameters X, Y, for which a sufficient number of points gives 20 to the considered temporal parameter (here the movement), a singular value which can be directly determined by the system or even be predetermined through API.
- the operation of the system can be presented in the following way:
- the bus 102 supplies the program calculation units 100 and 101 with the data which they process.
- the histogram calculation unit will supply through the line 103, a classification signal S M v ⁇ (i, j, t) which has the value 1 for all the pixels i, j which have a movement parameter around Vi, more precisely between the limits ViA, ViB determined by the histogram calculation unit 100 as a function of the API, S (i, j, t) taking the value 0 when the movement parameter has a value outside the interval defined by the above-mentioned terminals.
- the histogram calculation unit 101 receives this classification value of the movement and does not. therefore validates that the points X, Y for which the movement parameter MVT has a value close to Vi.
- the stogram calculation unit 101 supplies a classification value S ⁇ 5 ⁇ which supplies through line 104, possibly through bus 110, to the histogram calculation unit 100.
- Line 105 provides outside, for any use, possibly through the bus, the value of the parameter Vi and line 106 provides the classification value denoted here ZA which makes it possible to define an area whose contour
- FIG. 8 is approximately representative of all the points 203 animated by the speed Vi and its barycenter Xi,
- the system operates as indicated as long as a sufficient number of points determined by the API responds to the system, that is to say to a value of the movement parameter MVT comprised between the terminals ViA and ViB.
- the system is reset, again taking into account all the points in space 201 and therefore seeking a new characteristic area with regard to the values of the parameter MVT.
- the system processes all the points of space 201. If the histogram calculation units 100 and 101 isolate a number of. points having / values characteristic of the MVT parameter higher than a threshold, the classification signal ZA makes it possible to define a new area of interest (NEW ROI -Region Of Interest).
- a counter Cr is incremented by one unit per consecutive valid frame. This counter represents the confidence index of the tracking function. A new area is declared valid for a Cr value greater than 1.
- FIGS. 7 to 10 explains the operation resulting from the association of a histogram processing unit processing a time parameter with a processing unit processing space parameters. It is understood that in either of these categories, the parameters can be arbitrary. In particular, from the movement or speed parameter, the luminance, a hue or the spatial resolution can also be treated as a time parameter.
- the space parameters depend both on the space considered and on the frame used to define it.
- orthonorronnés landmarks of different orientations can be used, that allowing to define the best concentration of points for a given parameter being selected.
- This process of defining an area can be repeated and repeated in different ways.
- the representation Z will be used both to designate the classification signal associated with an area and the area itself.
- FIGS. 11 and 12 diagrammatically show how, from a main zone ZA associated with a parameter P, it is possible to obtain, for the same parameter P, sub-zones Za 0 , Za ⁇ , Za 2 .
- FIG. 11 the association of three histogram calculation units has been shown, each of which, according to criteria defined by the API, makes it possible to identify a sub-area identified, respectively Za o , Za ⁇ , Za 2 , from an area ZA.
- each sub-zone Za o can then serve as the main zone and allow the extraction of sub-sub-zones Za 00 , Za 0 ⁇ , Zao2 ...
- This process can be repeated as many times as desired by the operator and defined through the API to a degree of definition or fineness of the zones corresponding to the minimum number of points defined to constitute the threshold used in the prior description with reference to FIG. 10.
- Figure 13 is the representation of a hierarchical analysis of an object in the particular case of a face. It is therefore a particular application of the method and the device described above.
- the assembly represented in FIG. 7 makes it possible to identify a first area of interest delimited by the outline of the moving object which will be called the head area later. This area is characterized by a dominant color.
- another histogram calculating unit determines the variations or complements of colors inside this area and thus detects new homogeneous areas or sub-areas identifying the position of characteristic elements. of the face, such as the eyes and mouth.
- Each of these sub-areas is then analyzed separately in color, movement, shape, which allows them to highlight their specificity and to determine the face. .. '
- Each shape is advantageously analyzed from the spatial resolution, for example on the basis of the Gabor wavelet processing and their orientation.
- FIG. 13 there is shown the association of two histogram calculation units processing time parameters 301 processing the movement parameter and 302 processing a hue resulting from the association of the luminance, hue and saturation parameters such as shown in Figures 5 and 6.
- the Space Meter calculation units 303 to 305 allow the main area formed by the frame formed around the contour of the face in a certain number of sub-zones formed for example by the eyes, the mouth, ...
- the head can be defined at the first level by the dominant color of the face, hair, beard. For each of these characteristics, a higher spatial resolution indicates representative areas for a face, eyes and mouth, a third level would indicate by a finer spatial representation of the characteristics of the eyes.
- the description given so far corresponds to the use of the device of the invention for analyzing an object in space whose characteristics are not initially known.
- the device then adapts and identifies the characteristic zones by itself.
- the device of the invention can identify, among the data supplied to it, those corresponding to this definition. It is thus possible for him to locate and locate a predefined object.
- the invention relates to a method and an automatic collection device.
- the device includes a histogram calculation unit, preferably self-adapting, possibly with anticipation and learning. They are more particularly intended for the perception and processing of images.
- the selection criteria can be extremely varied. It can be a speed, a shape, a color ... or a combination of these criteria.
- Some of these methods and devices use a spatial and temporal processing unit which,. receiving a video signal S (PI),: produces a certain number of parameters for each pixel. These are for example the speed V, the direction DL, a time constant CO, and a binary validation parameter VL in addition to the delayed video signal VR and the various frame synchronization signals, line and pixel ' grouped under the name F.
- PI video signal S
- the object of the invention is to propose a method and a device for rapid and efficient automatic perception and to improve for such a device the units for constituting histograms by ensuring self-adaptation functions and, in. preferred embodiments of anticipation and learning. ' - •
- the present invention relates to a device for automatically perceiving an event occurring in a space with respect to at least one parameter.
- this device comprises a control unit, a data bus, a feedback bus and at least one histogram calculation unit for processing the parameter.
- the device comprises, for processing several parameters, several histogram calculation units organized in a matrix; ,.
- the histogram calculation units process data associated with pixels which together form a multidimensional space evolving over time and represented at a succession of moments, the data arriving at the calculation unit in the form of a DATA signal ( A) digital in the form of a sequence ay t of binary number of n bits associated with synchronization signals making it possible to define the given moment of space and the position of the pixel in this space, at which the signal has ijt received at a given instant • is associated, comprising:
- an analysis memory comprising a memory with. addresses, each associated with possible values, of numbers of n bits of the signal DATA (A). and the writing of which is controlled by a "WRITE" signal
- a classifier comprising a memory intended to receive a criterion C for selecting the DATA (A) parameter, receiving the signal DATA (A) as input and supplying, as output, a binary classification signal whose value depends on the result of the comparison of the DATA signal (A) with the selection criterion C
- a retroannotation unit receiving the output signal from the classifier and, from outside the histogram calculation unit, individual classification binary signals relating to parameters other than DATA (A), said retroannotation unit providing at output a positive global validation signal when all the individual retroannotation signals are positive,
- an address multiplexer - an increment validation unit, the counter of each memory address corresponding to the value d of a ⁇ at a given instant, being incremented by one when the retroannotation unit provides at output a positive global validation signal, the unit for calculating and storing statistical data exploiting at the end of the reception of the data ayt corresponding to the space at a time the content of the memory to update the unit analysis output, the memory being erased before the start of each frame for a space at a time by an initialization signal "INIT”, and in addition:
- the classifier memory is an addressable memory allowing the selection criterion to be updated in real time and having a DATA IN data input, an ADRESS address command and a WR write command, receiving on its input the output analysis memory and an END signal on its write command,
- the histogram processing units further include a data input multiplexer having two inputs and one output, receiving on one of its inputs a COUNTER counting signal and, on the other of its inputs , the data sequence a ⁇ and outputting the data sequence a- ⁇ to the address control of the classifier memory and an OR operator controlling the address multiplexer and receiving on its inputs an initialization signal INIT and the end signal END;
- the space is two-dimensional and the DATA signal (A) is associated with the pixels of a succession of images;
- the histogram processing units include means for anticipating the value of the selection criterion; - the means of anticipation of. the value of the selection criterion comprises memories intended to contain the values of statistical parameters relating to two successive frames;
- the statistical parameters are the average values of the validated agt data; . - the analysis output register of the histogram calculation units constitutes and keeps in memory at least one of the following values: the minimum value "MIN”, the maximum value "MAX”, the maximum number of pixels for which the signal V ijt has a particular value "RMAX”, the corresponding particular value POSRMAX, the total number of validated pixels "NBPTS";
- the statistical comparison parameter used by the classifier is RMAX 2.
- It comprises a controlled multiplexer, which can receive as input several statistical parameters and the nature of the comparison made by the classifier depends on the control of this multiplexer;
- certain histogram calculation units at least, comprise a learning multiplexer intended to receive an external control signal and producing an operation according to a learning mode in which the registers of the classifier and of the retroannotation unit are erased at the start of the exploitation of a frame and the analysis output register provides values characteristic of the sequence for each of these registers; . . .
- the memory of the classifier is made up of a set of independent registers each comprising an input, an output and a write command, the number of these registers being equal to number n of bits of the numbers of the sequence Vjjt, and it includes a decoder making it possible to output a write command signal corresponding to the associated input value (address) and a multiplexer controlled by this input value, making it possible to read the chosen register;
- at least some histogram calculation units include multiplexers, one of them being associated with the input of each register and. combinatorial modules connecting the registers, the multiplexers together
- the combinatorial modules include a morphological dilation operator comprising a logical unit "OR” with three inputs, the first of which receives the output signal from the rank register "Q", the second is connected to the output 0 of a logic unit “AND” with two inputs receiving respectively the output signal of the row register "Q + 1" and a positive expansion signal, the third is connected to the output of a logic unit “AND” with two inputs receiving respectively the output signal from the range register "Q-1" and a negative expansion signal; 5 '- in certain histogram calculation units at least, the combinatorial modules include a morphological erosion operator comprising
- each combinatorial modulated comprises a ... multiplexer, associating a morphological erosion operator and a morphological erosion operator; 0
- the invention relates to an automatic method perception of an event occurring in a space with respect to at least one parameter comprising digitize it and provide the input to a histogram calculation unit for forming a histogram representative of the parameter and deduce the desired result.
- the invention also relates to a method for analyzing a parameter representative of an event in an electronic device comprising a histogram calculation on data a ijt associated with pixels together forming a multidimensional space evolving in time and represented at a succession of moments, the data arriving at the calculation unit in the form of a digital DATA (A) signal in the form of a sequence a ijt of binary number of n bits associated with synchronization signals making it possible to define the given moment of space and the position of the pixel in this space, with which the signal ijt received at a given instant is associated in which - we associate with each data a ijt a binary classification signal whose value depends on the result of the comparison of the DATA signal (A) with the selection criterion C,
- FIG. 1 is a representation of the histogram calculation unit according to the invention, in its context
- FIG. 2 is a representation of the input video signal, processed by the device and method of the invention and the control signals generated by a sequencer;
- FIG. 3 is a diagram showing a passive histogram calculation unit
- FIG. 4 is a diagram representing a self-adaptive histogram calculation unit according to the invention with the anticipation and learning functionalities;
- FIG. 5 is a diagram representing signals used by. the calculation unit of FIG. 4;
- - Figure 6 is the flow diagram of the control software of the calculation unit of Figure 4 in master mode
- - Figure 7 is the flow diagram of the software for controlling the calculation unit of Figure 4 in slave mode
- FIG. 12 is a representation of the essential elements of the histogram calculation unit having a self-adaptation functionality
- FIG. 13 is a representation of a validation computer having several functions of self-adaptation;
- FIG. 14 is a representation of the elements of a histogram calculation unit producing POSMOY values;
- - Figure 15 is. a diagram representing the essential elements of ; . the self-adaptive histogram unit with anticipation - according to a first method; - Figure 15a is a representation similar to Figure 15 implementing a first generalized anticipation method;
- FIG. 16 is a diagram of the memory of the classifier
- FIG. 17 is a diagram representing the essential elements of the self-adaptive histogram unit with anticipation according to a second method
- FIG. 18 is a detailed representation of the memory of the classifier with an automaton, of elementary calculation per bit
- FIG. 22 is a representation of the retroannotation unit
- FIG. 23 is a block diagram of a logic reprogrammable unit • (FPGA) used as a retroannotation unit;
- FIG. 24 is the representation by register, limited to one row of the circuit, of Figure 23;
- - Figure 25 is a representation of the essential elements of a histogram calculation unit for learning
- - Figures 26 and 27 are schematic representations of a particular choice of axis
- FIG. 28 is a schematic representation of the • statistical display device
- FIG. 29 is an example of the result of the display produced by the device of Figure 28;
- FIG. 30 is the representation of the implementation of a plurality of histogram calculation units
- FIG. 31 is the representation of . the use of a programmable calculation unit single histogram with a multiplexer for its ⁇ for operating a plurality of parameters;
- FIG. 32 represents a set of histogram calculation units with programmable input control in their context of use constituting a functional unit; - '• " -' '
- FIG. 33 is a synthetic representation of a functional unit with the associated signal generator
- Figure 34 corresponds to Figure 32 in the case of a two-source acquisition
- - Figure 35 corresponds to Figure 33 in the case of a binocular acquisition
- - Figure 36 shows schematically a signal generator equipped with a controlled optics
- FIG. 37 shows the case of an acquisition with three sources; .
- - Figure 38 is a. representation of the management interface, of the application (API);
- - Figure 39 shows a device for processing a sound signal according to the invention;
- FIG. 40 is a simplified representation of a device according to the invention.
- the invention can be the subject of numerous embodiments.
- the information used can be varied in nature and represent multiple data or parameters, however, its first application is the processing of images, these constituting the space considered. It is of course, then, two-dimensional.
- the detailed description which follows corresponds to this particular embodiment.
- the histogram calculation unit 1 of the invention is represented in context by Figures 1 and 2.
- This histogram calculation unit 1 is part of a visual perception unit 13 which receives and uses a signal S (t) or S (PI).
- the histogram calculation unit uses and generates information known as retroannotation S '(t) on a bus 111. More specifically, FIG. 1 represents several histogram calculation units 1A, 1B, ..., 1 E associated in the same unit of visual perception.
- the visual perception unit 13 processes different signals relating to one or more visual scenes.
- the perception unit 13 processes other perception parameters, for example sounds, smells, etc. The description which follows mainly concerns visual perception, adaptation to. other parameters are possible.
- a sequencer 9 generates, from synchronization signals ST, SL, CLOCK, sequence signals INIT, WRITE and COUNTER which control the histogram calculation units.
- the input signals of the sequencer 9 can come from a signal generator assembly 2 comprising a camera 22 or from a signal generator assembly 3 comprising a CMOS imager 32.
- the sequencer 9 When the input signals come from a set 2 comprising a camera, this set imposes frame and line synchronization signals such that the histogram calculation unit and . sori sequencer 9 operate in slave mode or - slave synchronization -. On the contrary, in the case where these signals come from an assembly 3 comprising a CMOS imager, the sequencer 9 operates in master mode and generates the synchronization signals itself.
- the assembly 2 allows the acquisition of data coming from a scene 21 by a camera 22.
- the camera 22 produces a signal S (P1) whose shape, of the type shown in FIG. 2, will be described in detail below.
- the electronic control unit 23 of the camera 22 then supplies the signals S (t) resulting from the extraction of the synchronization signals from S (PI), ST, SL and the clock signal CLOCK from a loop to phase lock, which are used by the histogram calculation unit.
- this imager 32 is used for the acquisition of scene data 31, it supplies S (t) and is controlled by a synchronization unit 33 which produces the synchronization signals of frames ST and of line synchronization SL, as well as the clock signal CLOCK used both by the CMOS imager 32 and by the other elements of the visual perception unit 13.
- the histogram calculation units 1 are advantageously coordinated with a spatial 6 and temporal 5 processing unit which has been described in patent application WO-98/05002, and with a delay line 7.
- the spatial processing unit and temporal 5, 6 corresponds to the device referenced 11 in the cited patent application. It receives the signal S (Pl) and generates parameters V (speed), Dl (direction) each corresponding to one of the data identified by DATA (A) ... DATA (E) in the present application.
- This set consisting of a histogram calculation unit 1, the spatial and temporal processing unit 5, 6 and the delay line 7, provides either
- - video signal S - a signal which is composed of a succession of frames, each frame consisting of a succession of pixels the whole of which forms a space, for example an image for a two-dimensional space .
- the frames are themselves broken down into rows and columns.
- This signal S (t) carries a value ay of the parameter A for each of the pixels (i, j).
- the succession of frames therefore represents the succession of images over time.
- T represents the frame
- i is the number of a line in the frame T
- j is the number of the column of the pixel in this line
- a is the value of the parameter A associated with the pixel ijT .
- the signal S can be an analog signal. However, it is preferably digital and composed, as shown in FIG. 2, of a succession of frames Ti and T 2 , each being formed of a succession of scanned horizontal lines such as .i, I1.2, .. ., I 1 . 1 7 for T1 and l 2 . ⁇ ... for T 2 .
- Each line consists of a succession of pixels or PI image points.
- S (PI) comprises a frame synchronization signal (ST) at the start of each frame, a line synchronization signal (SL) at the start of each line which is not also a start of frame.
- S (PI) comprises a succession of frames which represents the time domain and, inside each frame, a series of lines and pixels in columns which are representative of the spatial domain.
- successive frames denote successive frames in time and "successive pixels at the same position” denote the successive values ay associated respectively with the pixels (i, j) placed in the same place in the successive frames , that is to say for example (1, 1) of ⁇ in the frame T 1 and (1, 1) of l 2 . ⁇ in the following corresponding frame T 2 ...
- the spatial 6 and temporal processing unit 5 From the S (PI), as indicated above with reference to the request PCT / FR-97/01354, the spatial 6 and temporal processing unit 5 generates one or more DATA (A) ... DATA signals (E).
- the passive histogram calculation unit without anticipation uses a DATA signal (A) whose structure is represented in FIG. 2.
- This signal can come either directly from a camera or from '' any image acquisition system, or have previously undergone a first treatment, for example a spatial and / or temporal treatment.
- the histogram calculation unit 1 comprises an analysis memory 100, an address multiplexer 105, a data input multiplexer 106, an increment validation unit 107, a classifier 101, a retroannotation unit 102 and a test unit 103 whose operations will be described later.
- All the elements making up the histogram calculation unit are controlled and synchronized by a clock signal (dock).
- This histogram calculation unit 1 includes an analysis memory 100.
- This analysis memory 100 is preferably a memory
- This memory has a number n of addresses d equal to the number of possible levels for the values of parameter A which must be 0 discriminated.
- Each of these addresses can store at least the number of pixels contained in a frame (that is to say in an image).
- the analysis memory 100 is capable of receiving the DATA signal (A). For each frame received, the pixels for which the value of the parameter A has a value a ⁇ d (if they are validated by the validation signal 102s which will be described later), increment the content of the rank address d of the memory 100 with a value 1. Thus, 0 after having received a complete frame, the memory 100 comprises, at each of its addresses d, the number of pixels which are validated and for which the parameter A has a value d . II. Address and data entry multiplexers
- the histogram calculation unit 1 also includes an address multiplexer 105, a data input multiplexer 106.
- Each of these multiplexers has a binary selection command, two inputs and one output.
- the output of the multiplexer has the value of one of the inputs when the selection command is worth 1 and the other when it is zero.
- the address multiplexer 105 selects an address in the analysis memory 100 as a function of the level d of the signal (a ljT ) received, the data input multiplexer 106 transfers l incrementation of the value contained in this memory according to the state of its selection command.
- the address multiplexer 105 transfers the signal from the counter which increments the address from zero to the maximum value of DATA (A).
- the data input multiplexer 106 forces zero on the input of memory 100.
- the histogram calculation unit also includes an increment unit 107. It is a controlled incrementer comprising an input, a command and an output.
- the output of the incrementing unit is equal to the output of the analysis memory 100 if the validation signal 102s is equal to zero, it is equal to this same value increased by 1 otherwise.
- the passive histogram calculation unit also includes a passive classifier 101 which includes a register 101r capable of storing some of the possible level values (di, d 2 , ...) of the levels of parameter A.
- the classifier 101 receives the DATA signal (A) and performs a sorting of the pixels providing, on its output 101s, a value 1 when the parameter A associated with said pixel has a level corresponding to that contained in the register 101r (di, d 2 , ...) and the value zero otherwise.
- the output of the classifier 101 is connected to a bus 111.
- the histogram calculation unit also has one. retro-annotation unit 102; : This retroannotation unit 102 is connected to the bus 111. It comprises at least one register 102r and receives for each pixel of the DATA (A) signals, the values (in E , ..., in B , Î ⁇ A ) of the outputs classifiers 101 of the different self-adaptive histogram calculation units connected to bus 111.
- the histogram calculation unit also includes a test unit
- analysis output registers 104 are intended to receive
- MIN minimum
- MAX maximum
- POSRMAX position of points for which information has been received.
- the test unit 103 updates the analysis output registers 104 according to the information it receives.
- the increment validation unit 107 also outputs a signal addressed to the test unit 103 allowing it to increment the analysis output register 104 in the favorable hypothesis.
- the analysis output registers 104 include memories for each of the key characteristics which include the minimum (MIN) of the histogram, the maximum (MAX) of the histogram, the number of points (NBPTS) of the histogram , the position (POSRMAX) of the maximum of the histogram and the number of points (RMAX) of the maximum of the histogram. These characteristics are determined in parallel with the formation of • the histogram by the test unit 103, as follows: For each pixel that is enabled:
- each of them supplies the bus, for each pixel, with the output signal 101s from its classifier 101 and they receive 'each of these signals on the input in Al ..., in E of their retroannotation unit 102.
- the value of the parameter, for example DATA (A) for the unit 1A, is compared with the content of the register 101r of the classifier 101.
- the result inA 101s of this comparison is a binary signal which is addressed at the same time as its counterparts inB ... inE from other units
- the retroannotation unit 102 compares these values taken together " with the content of its register 102r constituting a criterion R of retroannotation and generates, on its output 102s, a binary signal whose value depends on the result of the comparison.
- This signal 102s controls the incrementer 107. When it is equal to 1, it produces, by means of the data multiplexer 106, the incrementation of a unit of the content of the register of memory 100 corresponding to the value of DATA parameter (A), simultaneously the test unit 103 ensures the statistical exploitation of the content of the memory 100 and transfers the content to the analysis output register 104.
- the registers of the memory 100 each contain for value d, the number of pixels for which the signal DATA (A) had the corresponding value d and that the retroannotation unit has validated.
- the COUNTER signal which scans the values from 0 to n, ensures the reset of the registers of memory 100.
- the memory of the classifier 101 is a register 101r whose content determined from outside the system is fixed. This classifier is then said to be passive.
- Self-adaptation consists of the automatic updating, by the system itself, of the content of the memory of the classifier, this content then being a transcoding table (LUT - Look Up Table).
- a self-adaptive histogram calculation unit 1 To perform the self-adaptation function, that is to say updating the classifier in real time, the histogram calculation unit of FIG. 3 is improved in accordance with FIG. 4.
- the classifier 101 has an addressable memory whose writing is controlled by an END signal.
- the sequencer 9 generates this signal END shown in FIG. 5.
- the histogram calculation unit includes an OR selection operator 110 receiving the input signals INIT and END and connected at the output to the control of the address multiplexer 105.
- the memory of the classifier 101 is controlled by the system itself. Its content can be modified, it includes a DATA IN data entry, a WR write command and an ADRESS address entry.
- This address input is connected to the output of an anticipation multiplexer 108.
- This multiplexer 108 of the "two to one" type includes a anticipation command connected to the output of an OR operator 112 receiving the signals as an input INIT and END. .
- the inputs of the anticipation multiplexer receive the same signals as the inputs of the address multiplexer 105 (DATA (A) and COUNTER).
- the signal END is equal to 1
- the memory of the classifier is written by a signal resulting from the comparison between the value of the histogram memory 100 and a value coming from the analysis output register 104 (RMAX 2) for all the values DATA (A)).
- the classifier provides a classification function f A which is the relationship it establishes between the DATA (A) data it receives and the output value 101s it produces.
- the classifier 101 ensuring the self-adjustment includes a memory 118 whose write WR input receives the signal END, the 'Address ADDRESS input receives the output signal of the address multiplexer 108 It further comprises a comparator 119 comprising two inputs and an output, the latter being connected to the data input DATA IN of the memory 118.
- the first input of comparator 119 receives the value RMAX / 2 produced by an analysis output register 104 and, its second input receives the output of memory 100.
- each word has only one bit.
- a writing sequence begins.
- the value read is greater than RMAX / 2
- a value 1 is written in the memory 118 at the corresponding position. Otherwise, the value 0 is entered at this position.
- the set of memories d is scanned from 0 to n. Memory 118 . classifier 101 has been updated.
- Figure 1.3 shows an alternative embodiment of the classifier
- a multiplexer. 120 is controlled by a choice command 124 and allows the comparison of the parameter P with a statistical value Q, which can be developed in different ways depending on the parameters
- the input 0 of the multiplexer 120 receives the value RMAX / 2 produced from the data of the analysis output register 104 by the divider by 2, 121, the input 1 of the multiplexer 120 directly receives the value
- the input 2 of the multiplexer 120 receives a threshold value contained in a "THRESHOLD" register. 123 the content of which is programmed from outside . system, the input, 4 of this multiplexer receives the quotient of the number of NBPTS points by the THRESHOLD produced by the divider 122.
- the parameter P can be any parameter P. So as shown in Figure 13, the parameter P can be any parameter P.
- the content of the memory 118 is updated, as a function of the signals supplied by the comparator 119 in a manner similar to the update described in the first embodiment.
- FIG. 13bis and 13ter represents a third embodiment in which the sum of the occurrences in the histogram is used instead of the levels thereof.
- the classification limits are defined for example by the use of an RMAX- register, corresponding to the maximum number of occurrences of the parameter analyzed, by searching for the values of the parameter for RMAX / 2. On either side of the position of RMAX, these values correspond to terminal A and terminal B of the classifier.
- the RMAX register is replaced by the NBPTS register, corresponding to the total accumulation of occurrences, and to deleting a percentage k of NBPTS on either side of the histogram. Terminals A and B become more stable.
- a device can easily perform this function.
- a first test consists in assigning to Terminal A, the value of increment i as long as Si is smaller than the value ⁇ previously defined.
- a second test consists in assigning to Terminal B, the value of increment i as long as Si is smaller than the value ⁇ previously defined.
- FIG. 16 is a detailed representation of the memory 118 showing a demultiplexer with input validation 130 and a multiplexer
- the input demultiplexer 130 receiving the write signal
- the output multiplexer 131 addresses the value of a particular register, selected by the address command ADRESS, on the output 101s of the memory 118 of the classifier.
- the input demultiplexer 130 and the output multiplexer 131 are controlled by the bus 134 coming from the anticipation multiplexer 108.
- the 1 / n input demultiplexer 130 controlled by the address transmitted by the bus 134, sends the signal WR (WRITE), respectively in the form of the signals Sel 0 , Seli, Sel 2 , ..., Breast on the registers 140 o , 14O 1 ( 140 2 140 n of rank 0, 1 ..., n and determines which of these registers in which the content of the information transmitted by the signal DATA IN is addressed.
- the information coming from these registers 140 o , 140 ⁇ , 140 2 , ..., 140 n are addressed to the multiplexer 131 which directs them to its output OUT.
- the histogram constitution unit 1 performs a anticipation function.
- the aim of anticipation is, as its name suggests, to anticipate the value contained in the memory 118 of the classifier so as to speed up processing and, thereby, to facilitate the monitoring of an object or its evolution .
- anticipation defines a anticipation function which, combined with the classification function f, produces a function (fog) A connecting the DATA data (A) and the output value (101s) A characterizing the histogram processing unit processing parameter A.
- POSMOY is the value of the DATA (A) parameter with respect to which, in a given frame, said parameter has a value greater than or equal to half of the validated points of the frame and a lower value for the other half.
- This POSMOYo variable is produced by a comparator 302.
- This comparator 302 receives, on one of its inputs Q, the parameter NBPTS which is divided by two by the divider 303. Its second input P is supplied by the output of a register 301 which is controlled by the initialization signals INIT and END END, and receives as input the output of an adder 300 which itself receives as input, the output value of register 301 and on its second input, the output value of memory 100 which was previously described.
- the register 301 initially reset to zero, stores the cumulative content of the registers of the memory which are scanned by the signal COUNTER from zero to n. •
- POSMOYo therefore contains the last COUNTER value for which the total is less than NBPTS / 2.
- This first method is represented in FIG. 15.
- the memory 118 is that described above with reference to FIG. 16.
- An automaton 310 said - unit for calculating absolute value with sign extraction - supplies the values jPOSMOYo - POSMOY ⁇
- FIG. 15a represents an alternative circuit allowing the implementation of the first method of applying the variation of the histogram to anticipation.
- the multiplexer 127 receives as input the two POSMOY functions, that is to say k1 x (P 0 - P + d and k2 x (P 0 - Pi) + c2, and provides a grounded output on the value of the "CLOCK" control signal.
- the OR circuit 125 and the delay circuit 126 can be used.
- the delay circuit is controlled by the same clock signal which controls the multiplexer 127.
- the output values of the memory 118 connected to the two different offset functions are then supplied to the OR gate 125 whose output is the signal 102s with an improved classification area, which improves anticipation.
- This second method is represented in FIG. 17.
- the memory 118 is then that represented in FIG. 18.
- the register 140 ⁇ is associated with a multiplexer of inputs 2/1 160 1 which receives on one of its inputs, the binary signal leaving the comparator 119 and on the other of its inputs, the output signal from the anticipation calculator 150-].
- the input multiplexer 160 ⁇ is controlled by the ETD signal which also controls the writing; '• • •
- the command to write to register 14 ⁇ ! is connected to an OR operator 170- 1 which receives on one of its inputs, the ETD signal and the other a signal Sel 0.
- an anticipation calculator 150 ⁇ receives as input the three signals Q 0 , Qi and Q 2 from the output of the registers 140 0 , 140- t , 140 2 of rank respectively 0, 1, 2, it is controlled by signals SM, SP and T.
- the anticipation calculator 150 is described in detail in FIG. 19, it comprises a multiplexer 207 comprising an output and two inputs controlled by the signal T.
- the expansion operator 208 comprises an OR circuit 201 with three inputs and one output. Its output is connected to multiplexer 207.
- the erosion operator 209 has an AND circuit 204 with three inputs and one output. Its output is connected to multiplexer 207.
- the first input is supplied by the Qi signal. Its second input is connected to a NAND circuit with four inputs and an output 205. The first input of this NAND circuit 205 is connected to the signal SP, the second to the signal Qi. The third input is connected to signal Q 0 and the fourth inverted to signal Q 2 .
- a second NAND operator 206 has four inputs and an output connected to the third input of the AND circuit 204, the first of its inputs being supplied by the signal Q 1 ( the second by the signal SM, the third by the signal Q 2 and the fourth reversed by the signal Q 0 .
- the signal INIT generated by the sequencer 9 starts the processing cycle of a frame. During its duration, all memories and registers are initialized.
- the WRITE signal also generated by the sequencer 9 follows the INIT signal and controls the statistical calculations for the frame considered, the data of which are represented by the curve C, the axes of which represent the values of the parameter on the abscissa and the number of occurrences on the ordinate. .
- the test unit 103 searches for the maximum number of RMAX occurrences. At the end of the WRITE signal, the signal END still generated by the sequencer 9 validates the updating of the memory of the classifier 118. The new data is generated by the comparator 119.
- the content of the memory 118 is represented by the distribution R 0 .
- the end of the signal END starts the signal ETD, the duration of which is determined by the command generator 313. This signal ETD validates the calculation of the extent in the memory 118 of the classifier.
- the respective durations of SP and SM determine the extent and the position of the distribution R5 at the end of the ETD signal.
- the multiplexer 207 controlled by the command T having two inputs supplied respectively by the outputs of the expansion and erosion operators and an output makes it possible to use one or the other of these operators as a function of the command T.
- the output of multiplexer 207 is OUT-i
- the retroannotation block 102 comprises a single register containing a single retroannotation value constituting the retroannotation criterion R.
- the retroannotation block is a memory which can contain several values together forming the criterion R of retroannotation, one or the other of which can validate the information carried by a pixel. Each of these values is stored in a product term register 410.
- FIG. 22 represents such a retroannotation block 102 as a whole. It is made up of a set of product terms 410 supplied by bus 425 A "PRODUCT TERM" and controlled by the Programm Register bus 12.
- Each of these product terms 410 has an output which feeds an OR operator 421 which itself outputs an incoming signal on one of the inputs of a controlled inverter 422, which receives on its second input the signals coming from the Programm bus Register 12 via register 423.
- FIGS 23 and 24 illustrate a reprogrammable memory unit (FPGA - Field Programmable Gâte Area) 400 implemented for the realization of the retroannotation unit 102.
- a memory comprises a controlled inverter 403 whose output is the output of the reprogrammable unit 400 and one of whose inputs is an OR operator 401 connected to the B lines 405, these lines intersecting the A columns 406 which are connected to.
- amplifiers 402 supplying signals s and. s ..
- intersections 404 of lines 405 and columns 406 are programmable connections making it possible to determine the entire operation of the reprogrammable unit 400.
- FIG. 24 represents a single line 410 of such a reprogrammable unit 400.
- Such a line comprises registers 411 and 412 intended to receive programming variables Reg-aO and Reg-bO.
- This line can be broken down into A elementary functions, each of which comprises a controlled inverter 413, an OR operator 415 and an inverter 414.
- One of the inputs of the controlled inverter 413 is connected to input A and the other of these entries in bit i of the Reg-aO register.
- This controlled inverter 413 is connected to the input of the inverter 414. which supplies, by its output, one of the inputs of the operator.
- the output of the controlled inverter 413 also supplies one of the inputs of an OR operator 417 which still receives all of the corresponding signals produced by the various outputs of the elementary functions.
- An AND operator 416 whose output is product termO receives as input on the one hand the output of the operator OR 417 and on the other hand the outputs of the various elementary functions.
- the reverse annotation block can be programmed from the outside by an instruction from an application management interface. This interface loads registers 411 and 412.
- the unit 1 of constitution. histogram in addition to updating the classifier and anticipation has a learning function.
- the histogram constitution unit 1 comprises a learning multiplexer 108 which allows automatic programming of the retroannotation unit 102.
- the learning multiplexer 108 selects one or the other of the two possible operating modes (operation and learning). In operating mode, the values contained in the register of the retroannotation unit 102 are fixed, on the contrary, in learning mode, these values are updated.
- the operating mode During operation in operating mode, the learning multiplexer 109 emits, on its output, a signal of value 1 signifying that the values contained in the registers of the retroannotation block 102 are not modified during the operating sequence in operating mode.
- the values stored in these registers were therefore either chosen and stored by the user, or were the result of a prior learning phase which we will analyze below.
- the retroannotation unit 102 also receives, from the other histogram calculation units cooperating with that described here, comparable signals inE ... inA.
- this retroannotation unit compares the values thus received with the values stored in its memory or in its register (s) and produces as output a signal 102s equal to 1 in the event of coincidence and to zero in the opposite case.
- This validation signal is sent to the increment validation unit and when its value is equal to 1, authorizes the taking into account of the value of the parameter DATA (A) of the pixel concerned in the analysis memory 100 and, otherwise, authorizes the processing of the next pixel.
- the learning mode The operation of the histogram calculation unit is then controlled by signals represented in FIG. 5, that is to say an initialization signal (INIT), a write signal (WRITE), which carry the information corresponding to each pixel of the frame (or of the image) and an end signal END.
- the learning multiplexer 109 outputs the value of the feedback signal which is then used as DATA (A).
- a signal LEARNj validates during a whole sequence the processing in learning mode.
- the training registers 116 are updated.
- the retroannotation block 102 ensures the transparency of the signals, validating the datum DATA (A), then equal to the retroannotation signal 111 as soon as at least one of the inputs inA, ..., in E is active.
- the histogram memory 100 represents the distribution of the retroannotation signal.
- the test unit 103 then produces a classification of occurrences by decreasing value in number equal to B "SUM TERM".
- the values of the feedback signal thus selected are written to registers 411 and 412 of each of the blocks 410 of the feedback block 102.
- Register 412 corresponds to the value of the feedback signal and register 411 corresponds to its complement. In practice, two outputs from the same register can be used providing these two values.
- learning requires, for n input parameters, n + 1 histogram calculation units.
- the n blocks processing one of the parameters and the remaining block processing the feedback information to ensure learning.
- the retroannotation information having a large number of digits, the learning unit is dedicated and is large.
- the spatial processing unit 6 produces at the output signals F, SR, V, VL, Dl, CO each associated with each pixel synchronously. They are preferably digital signals.
- the complex signal F then comprises a certain number of output signals generated by the system preferably comprising signals signifying the presence " and the location of an area or a moving object, the speed V is the direction of the movement DL of each pixel of the image.
- an output of the system is constituted by the digital input video signal which is delayed (SR) so as to synchronize it with the output ZH of the frame, taking into account account account the calculation time of the composite data signal F (for one frame).
- the delayed signal SR is used to represent the image received by the camera on a monitor or. a screen, television, which may also be used to representing the information contained in the composite signal ZH
- the composite signal ZH can also be transmitted to a separate processing unit 10a which continues the processing of this signal.
- the position of a pixel in space is represented relative to a system of axes. Depending on the shape, orientation, ... of the objects in the scene, certain axis systems provide better results than others.
- FIGS. 26 and 27 show the method for choosing the axes making it possible to obtain optimized histograms, that is to say having a particularly marked maximum.
- the Space transform unit receives as input the spatial data x, y which can be either Cartesian data or polar data.
- This Space transform unit is controlled by a signal ⁇ and provides at the output, for each of the values of ⁇ , a parameter which supplies a histogram constitution unit according to T invention.
- the command program for this histogram unit introduced by the Programm Register allows you to select the value ⁇ producing an optimized histogram.
- Binocular which by a controller for calculating distance provides a depth parameter.
- a curve generator 114 allows the overlay on the screen of a curve of the DATA values for the frame previously processed. Likewise, an overlay of the retroannotation signal is possible.
- FIGS. 28 and 29 describe more precisely the means for displaying the histogram curve.
- the memory 100 addressed by the value of the column counter 353 supplies one of the inputs of a shift register 350, the other input of which is supplied by the parameter RMAX produced by the analysis register 104.
- This shift register 350 feeds one of the inputs of a comparator 351, the other input of which is supplied by a line counter
- the display control block 365 of the screen 8 receives the delayed video signal SR, a cursor command produced by the cursor block 366, a command produced by the semi-graphic memory 367.
- FIG. 29 is the result of the sticker obtained 360 validated by the switch 16 which transfers the curved display signal to the overlay command 15 in the screen 361 which further comprises a command zone 362, a cursor 363 and a text area 364.
- this screen and the associated mouse constitute a graphical user interface (GUI) allowing the user to generate and control the application.
- GUI graphical user interface
- the retroannotation function can be viewed, dynamically, in the form of pixels 365, by switching the switch 17, on the command for overlaying the retroannotation 14 .:
- FIGS. 30 and following the implementation of a set of histogram calculation units has been shown, allowing the management of as large a number as necessary of parameters A, B, C, D, E ...
- the combination of both spatial (in principle two) and temporal (at least one) parameters makes it possible to model a spatio-temporal neuron.
- the temporal 5 and spatial 6 respectively processing units. , receive the signal on the one hand.
- S (t) on the other hand the clock signals.
- each of the parameters ⁇ , B, C, D, E ... leaving these temporal 5 and spatial processing units 6 feeds a histogram calculation unit, respectively 1 A , 1 B .-- 1 E -
- the retroannotation produced by all the classifiers is available on bus 111 and used as a whole by each of the histogram calculation units, respectively 1 A , 1 B ... 1 E-
- A, B, C, ..., E can respectively represent the color components of the input pixel in the form of luminance L, hue T and 'saturation S.
- D and E would be the coordinates Pi and P 2 of the pixel considered in an optimized axis system.
- each histogram calculation unit 1 A , 1 B , 1 E processes one of the data DATA (A), DATA (B), ... DATA (E) by the corresponding function (fog) A ... to individually produce an output value (101s) A ... and all together the retroannotation available on the bus 111.
- the choice of the parameter processed by each histogram calculation unit as well that the fog function it performs, are determined by the API
- the various parameters DATA (A) ... DATA (E) supply an input multiplexer 500 which is controlled by a register 501.
- the register 501 is updated by the command SELECT referenced 502.
- the learning controlled multiplexer 503 receives, depending on the state of the learning command of the histogram calculation unit i, LEARNi, either the retroannotation information transmitted by the bus 111, or the information from the input multiplexer 500. It is thus possible to use a single histogram calculation unit 1 to process any of the different parameters A, B, C, ..., E which are sent to it by a bus 510 according to the SELECT command.
- the histogram calculation unit will operate either in operation or in learning.
- the assembly 1a thus formed by a histogram calculation unit 1, an input multiplexer 500, its associated register 501, and possibly a learning multiplexer constitutes a versatile histogram calculation unit.
- FIG. 32 represents a complete device comprising, by way of example, a set of sixteen versatile histogram calculation units.
- Bus 111 carries retroannotation information.
- the entire application is controlled by the control unit 513 which determines which of the parameters L, T, S, N, D, p 0 , Pi ... p-is which are processed at a given time by a or a group of dedicated multipurpose histogram units and, by the sequencer. 9.
- FIG 40 shows a block diagram of an assembly comprising several histogram calculation units (as shown in Figure 31) in accordance with one embodiment of the present invention.
- Each histogram calculation unit 1a is connected to a data bus 510 which supplies the various parameters to be processed, and to a bus 11 which supplies the classification signal 101s and the learning function signals to the various units 1a.
- Each histogram calculation unit comprises a memory 100, a classifier 101 and a retroannotation unit 102.
- Each unit 1a is capable of having functions of automatic classification of anticipation and learning as described above.
- the set of histogram calculation units 1 can operate in processing mode while one or more of them are in learning mode
- a calculation unit: of histogram is used in timeshare by several parameters ' during each frame, possibly stored in a memory not shown.
- the unit of. histogram calculation 1 calculates histograms and the corresponding statistics for two or more parameters (for example DATA (A) and DATA (C)) during each frame.
- a multiplexer 500 provided in this embodiment is capable of multiplexing the different parameters. In this way, one . a limited number of histogram calculation units is necessary for processing a greater number of parameters, which makes it possible to reduce the amount of silicon necessary to manufacture the useful number of histogram calculation units.
- the generic visual perception processor (GVPP) 520 thus formed can be integrated on a single semiconductor substrate.
- the number of versatile histogram calculation units 1a depends on the application and on the technologies for manufacturing semiconductor components available.
- the 0.5 ⁇ m technology currently available allows the integration of 32 units economically. With advanced semiconductor techniques, it becomes impossible to manufacture more and more histogram calculation units (for example the blocks 1a of FIG. 32) on the same component and to carry out more numerous calculations on more samples (i.e. increasingly large numbers of sampling per
- Such an increase in computing capacity can be achieved without increasing the complexity of the API, which is presented later and shown in detail in Appendix A.
- the same set of instructions can command a set of 20 units as a set of 200 or even 2000 units without increasing its complexity.
- a calculation unit 605 according to the present invention, that is to say similar to the calculation unit 520 of FIG. 32, is used to process the associated parameters to a domain of perception other than the visual domain.
- the technique of the present invention is applied to oral or sound analysis, for example for speech recognition or the use of voice to enter text into a computer.
- a device generating an audible signal supplies an audible signal to the computer 605 which
- the device generating the signal includes a microphone 600 but can also include any device capable of providing analog or digital signals, for example a CD or DVD player ...
- the signal generating device provides preferably digital signals and can operate in a slave mode or in a master mode in the same way as the signal generator assembly 2 of FIG. 1.
- the computer 605 receives the signals and processes various parameters of the sound signal. These parameters include frequency, amplitude and phase. The phase and amplitude parameters are respectively analogous to the spatial and temporal signals used for the processing of visual scenes.
- the computer 605 supplies signals to the device 610 so as to allow the representation of the results.
- the device 610 includes a printer allowing the printing of texts associated with the signals supplied by the signal generator 600.
- the device 610 can include a monitor or any other text generating device.
- FIG. 33 is the representation of a generic visual perception processor 520 or 530 receiving information from a CMOS imager 521 comprising a retina 522 and a sequencer 523.
- FIG. 34 represents a complete system capable of constituting a complete application and operating with several CMOS imagers.
- FIG. 37 schematizes a system made up of a set of polyvalent histogram calculation units capable of processing information coming from. Three directions, respectively V1, V2 and V3 which can represent a three-dimensional space. We can thus manage volume perception data and use it in the field of robotics.
- the application management interface (API - Application Programm Interface) represented in FIG. 38 makes it possible to provide a complete system or generic visual perception processor composed of a number of versatile histogram calculation units, all the external parameters it needs. This ensures its dynamic configuration.
- the annex A which is attached, is an integral part of the description of the present application, it provides a functional diagram of the spatiotemporal unit API, the graphical user interface (GUI) API, the mouse API and the I / O API , as well as the various associated API commands.
- GUI graphical user interface
- Each of the command mnemonics is associated with an index i corresponding to the number of the multipurpose histogram calculation unit for which it is intended. It can be accompanied by configuration parameters. It allows you to assign the DATA (A) ... DATA (E) parameters to real parameters of the observed scene.
- SELECT allows you to assign a DATA (A) parameter to a specific unit. LEARNj makes it possible to carry out learning for a unit of polyvalent histogram calculation i. fiSTART ensures the initialization of a versatile histogram calculation unit. This command configures the memory 118 of classifier 101.
- STOP ensures that the multipurpose histogram calculation unit is stopped. It is used as soon as a histogram calculation unit is inactive. This reduces the energy consumption of the whole.
- AFCURV is the curve display command controlling the switch 16 shown in FIG. 4. Its reverse command is CLCURV.
- AFMAP is the retroannotation display command controlling switch 17. Its reverse command is CLRMAP.
- MAP is the write command, of registers 411 and 412 of the retroannotation unit 102:.
- GETLRN is the command ensuring the recovery of the content of the retroannotation registers 411 and 412 after learning.
- API Application Programming Interface
- the interface has 4 subdivisions for GVPP:
- Block3 equ 03 equivalence Block3 is equal to the value 3
- Role Selection of the input signal of a block. For example luminance, hue, saturation, orientation of lines in the plane ... etc. Parameters: Number of the block concerned, signal to be selected. totype:
- Input - RO Block number.
- R1 . Input parameter
- Parameters Number of the block concerned, parameter (s) to retrieve.
- Role ' Switching of a block in learning mode.
- Role Programming of the block according to a previous learning to change context. Search for another event or object: write the block annotation matrix. Parameters: Number of the block concerned, Logical combination of the other associated blocks; sum of terms produced (AND and OR).
- Role Clear the screen of the block's annotation table.
- Parameters Data to display and position on the screen.
- Role Generate a click cursor.
- MVCAM Inlet - RO X position R1 Y position R2: Focus Outlet -
- Role Send information to the PC.
- Parameters point to information and number of information.
- Role Recover information from the PC.
- FIG.3 WRITE
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
Description
Claims
Priority Applications (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2002568307A JP2004523047A (ja) | 2001-02-23 | 2002-02-22 | 対象物をその形状、そのサイズ及び/又はその方向によって認知する方法及び装置 |
| CA002439867A CA2439867A1 (fr) | 2001-02-23 | 2002-02-22 | Procede et dispositif de reperage d'un objet par sa forme, sa dimension et/ou son orientation |
| EP02716806A EP1364341A1 (fr) | 2001-02-23 | 2002-02-22 | Procede et dispositif de reperage d'un objet par sa forme, sa dimension et/ou son orientation |
| MXPA03007554A MXPA03007554A (es) | 2001-02-23 | 2002-02-22 | Metodo y dispositivo para la percepcion de un objeto por su forma, su tamano y/o su orientacion. |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| FR0102539 | 2001-02-23 | ||
| FR0102539A FR2821459B1 (fr) | 2001-02-23 | 2001-02-23 | Procede et dispositif de reperage d'un objet par sa forme, sa dimension et/ou son orientation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2002069265A1 true WO2002069265A1 (fr) | 2002-09-06 |
Family
ID=8860406
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/EP2002/001899 WO2002069265A1 (fr) | 2001-02-23 | 2002-02-22 | Procede et dispositif de reperage d'un objet par sa forme, sa dimension et/ou son orientation |
Country Status (6)
| Country | Link |
|---|---|
| EP (1) | EP1364341A1 (fr) |
| JP (1) | JP2004523047A (fr) |
| CA (1) | CA2439867A1 (fr) |
| FR (1) | FR2821459B1 (fr) |
| MX (1) | MXPA03007554A (fr) |
| WO (1) | WO2002069265A1 (fr) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018197693A1 (fr) * | 2017-04-28 | 2018-11-01 | Another Brain | Procédé et dispositif automatisés aptes à assurer l'invariance perceptive d'un évènement spatio-temporel dynamiquement en vue d'en extraire des représentations sémantiques unifiées |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR2843467B1 (fr) * | 2002-08-07 | 2005-03-25 | Holding Bev Sa | Procede et dispositif a recrutement dynamique de modules de calcul et traitement d'histogrammes pour caracterisation et reconnaissance d'objets |
| AU2003282816A1 (en) * | 2002-08-07 | 2004-02-25 | Holding B.E.V. Sa | Active visual perception method and device for characterisation and recognition through the analysis of mono/multidimensional parameters in multiclass computing units and histogram processing, dynamic unit recruitment |
| FR2858447A1 (fr) * | 2003-07-29 | 2005-02-04 | Holding Bev Sa | Procede et dispositif automatise de perception avec determination et caracterisation de bords et de frontieres d'objets d'un espace, construction de contours et applications |
| FR2884625A1 (fr) * | 2005-04-15 | 2006-10-20 | Holding Bev Sa Sa Luxembourgeo | Procede de guidage automatique de vehicule dans une voie de circulation, dispositif correspondant |
| JP4696991B2 (ja) * | 2006-03-22 | 2011-06-08 | 日産自動車株式会社 | 動き検出方法および動き検出装置 |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR2611063A1 (fr) * | 1987-02-13 | 1988-08-19 | Imapply | Procede et dispositif de traitement en temps reel d'un flot de donnees sequence, et application au traitement de signaux video numeriques representatifs d'une image video |
| US5795306A (en) * | 1994-03-10 | 1998-08-18 | Mitsubishi Denki Kabushiki Kaisha | Bodily state detection apparatus |
-
2001
- 2001-02-23 FR FR0102539A patent/FR2821459B1/fr not_active Expired - Fee Related
-
2002
- 2002-02-22 CA CA002439867A patent/CA2439867A1/fr not_active Abandoned
- 2002-02-22 WO PCT/EP2002/001899 patent/WO2002069265A1/fr not_active Application Discontinuation
- 2002-02-22 EP EP02716806A patent/EP1364341A1/fr not_active Withdrawn
- 2002-02-22 JP JP2002568307A patent/JP2004523047A/ja active Pending
- 2002-02-22 MX MXPA03007554A patent/MXPA03007554A/es unknown
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR2611063A1 (fr) * | 1987-02-13 | 1988-08-19 | Imapply | Procede et dispositif de traitement en temps reel d'un flot de donnees sequence, et application au traitement de signaux video numeriques representatifs d'une image video |
| US5795306A (en) * | 1994-03-10 | 1998-08-18 | Mitsubishi Denki Kabushiki Kaisha | Bodily state detection apparatus |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018197693A1 (fr) * | 2017-04-28 | 2018-11-01 | Another Brain | Procédé et dispositif automatisés aptes à assurer l'invariance perceptive d'un évènement spatio-temporel dynamiquement en vue d'en extraire des représentations sémantiques unifiées |
| FR3065825A1 (fr) * | 2017-04-28 | 2018-11-02 | Patrick Pirim | Procede et dispositif automatises aptes a assurer l'invariance perceptive d'un evenement spatio-temporel dynamiquement en vue d'en extraire des representations semantiques unifiees |
| EP3616132A1 (fr) * | 2017-04-28 | 2020-03-04 | Another Brain | Procédé et dispositif automatisés aptes à assurer l'invariance perceptive d'un évènement spatio-temporel dynamiquement en vue d'en extraire des représentations sémantiques unifiées |
| US11164049B2 (en) | 2017-04-28 | 2021-11-02 | Another Brain | Automated method and device capable of providing dynamic perceptive invariance of a space-time event with a view to extracting unified semantic representations therefrom |
Also Published As
| Publication number | Publication date |
|---|---|
| FR2821459B1 (fr) | 2004-11-12 |
| FR2821459A1 (fr) | 2002-08-30 |
| MXPA03007554A (es) | 2003-12-04 |
| CA2439867A1 (fr) | 2002-09-06 |
| EP1364341A1 (fr) | 2003-11-26 |
| JP2004523047A (ja) | 2004-07-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hernandez-Ortega et al. | Deepfakeson-phys: Deepfakes detection based on heart rate estimation | |
| EP1259939B1 (fr) | Procede et dispositif de perception automatique | |
| CN104246656B (zh) | 建议的视频编辑的自动检测 | |
| Kim et al. | Exposing fake faces through deep neural networks combining content and trace feature extractors | |
| US12217474B2 (en) | Detection of moment of perception | |
| Hernandez-Ortega et al. | Deepfakes detection based on heart rate estimation: Single-and multi-frame | |
| Planamente et al. | Da4event: towards bridging the sim-to-real gap for event cameras using domain adaptation | |
| WO2018197693A1 (fr) | Procédé et dispositif automatisés aptes à assurer l'invariance perceptive d'un évènement spatio-temporel dynamiquement en vue d'en extraire des représentations sémantiques unifiées | |
| WO2022245399A1 (fr) | Utilisation d'une interpolation pour générer une vidéo à partir d'images statiques | |
| CN118101856B (zh) | 图像处理方法及电子设备 | |
| Zhang et al. | Face spoofing video detection using spatio-temporal statistical binary pattern | |
| WO2002069265A1 (fr) | Procede et dispositif de reperage d'un objet par sa forme, sa dimension et/ou son orientation | |
| Kannoth et al. | Hand Gesture Recognition Using CNN & Publication of World's Largest ASL Database | |
| WO2020193972A1 (fr) | Analyse faciale | |
| Menon et al. | Style transfer applied to face liveness detection with user-centered models | |
| Pundir et al. | A review of deep learning approaches for human gait recognition | |
| WO2002075659A1 (fr) | Procede et dispositif d'adaption de la sensibilite de detection d'un parametre porte par un signal | |
| Dalal et al. | False media detection by using deep learning | |
| Bhamare et al. | Translating the unspoken Deep learning approaches to Indian Sign Language recognition using CNN and LSTM networks | |
| Mocanu et al. | Gesture-based human-robot interaction | |
| US12254406B2 (en) | Object observation prediction in images using encoder-decoder models | |
| Murtaza et al. | Face forgery detection via optimum deep convolution activation feature selection algorithm using expert-generated images | |
| Bansal et al. | Detection of Anomalies in Surveillance Video using Improved Cheetah Optimizer Algorithm along with Variational Auto Encoder Classifier | |
| Wen | Reconstruction of human faces from voice | |
| Pola et al. | An Effective Deep Learning Model for Hand Shadow Puppet Recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| AK | Designated states |
Kind code of ref document: A1 Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NO NZ OM PH PL PT RO RU SD SE SG SI SK SL TJ TM TN TR TT TZ UA UG UZ VN YU ZA ZM ZW |
|
| AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): GH GM KE LS MW MZ SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG |
|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
| WWE | Wipo information: entry into national phase |
Ref document number: 2002716806 Country of ref document: EP |
|
| WWE | Wipo information: entry into national phase |
Ref document number: PA/a/2003/007554 Country of ref document: MX Ref document number: 2002568307 Country of ref document: JP Ref document number: 2439867 Country of ref document: CA |
|
| DFPE | Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101) | ||
| WWP | Wipo information: published in national office |
Ref document number: 2002716806 Country of ref document: EP |
|
| REG | Reference to national code |
Ref country code: DE Ref legal event code: 8642 |
|
| WWW | Wipo information: withdrawn in national office |
Ref document number: 2002716806 Country of ref document: EP |