+

WO2000045339A1 - Appareil et procede de description des parametres de mouvement d'un objet dans une sequence d'images - Google Patents

Appareil et procede de description des parametres de mouvement d'un objet dans une sequence d'images Download PDF

Info

Publication number
WO2000045339A1
WO2000045339A1 PCT/US2000/002364 US0002364W WO0045339A1 WO 2000045339 A1 WO2000045339 A1 WO 2000045339A1 US 0002364 W US0002364 W US 0002364W WO 0045339 A1 WO0045339 A1 WO 0045339A1
Authority
WO
WIPO (PCT)
Prior art keywords
image sequence
trajectory
interval
motion
optical flow
Prior art date
Application number
PCT/US2000/002364
Other languages
English (en)
Inventor
Edmond Chalom
Sriram Sethuraman
Iraj Sodagar
Original Assignee
Sarnoff Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sarnoff Corporation filed Critical Sarnoff Corporation
Publication of WO2000045339A1 publication Critical patent/WO2000045339A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/783—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G06F16/7847—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using low-level visual features of the video content
    • G06F16/786—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using low-level visual features of the video content using motion, e.g. object motion or camera motion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/49—Segmenting video sequences, i.e. computational techniques such as parsing or cutting the sequence, low-level clustering or determining units such as shots or scenes

Definitions

  • the invention relates to image processing for describing the motion of object(s) in image sequences, e.g., video. More particularly, the invention relates to an efficient framework for object trajectory segmentation, which in turn, can be employed to improve image processing functions, such as context-based indexing and retrieval of image sequences with emphasis on motion description. BACKGROUND OF THE DISCLOSURE
  • MPEG-7 Moving Picture Experts Group
  • typical content description of a video sequence can be obtained by dividing the sequence into “shots".
  • a "shot” can be defined as a sequence of frames in a video clip that depicts an event and is preceded and followed by an abrupt scene change or a special effect scene change such as a blend, dissolve, wipe or fade. Detection of shot boundaries enables event-wise random access into a video clip and thus constitutes the first step towards content search and selective browsing.
  • representative frames called "key frames” are extracted to capture the evolution of the event, e.g., key frames can be identified to represent an explosion scene, an action chase scene, a romantic scene and so on. This simplifies the complex problem of processing many video frames of an image sequence to just having to process only a few key frames.
  • the existing body of knowledge in low-level abstraction of scene content such as color, shape, and texture from still images can then be applied to extract the meta-data for the key frames.
  • Motion information can considerably expand the scope of queries that can be made about content (e.g., queries can have "verbs” in addition to "nouns"). Namely, it is advantageous to have additional conditions on known information based on color, shape, and texture descriptors, be correlated to motion information to convey a more intelligent description about the dynamics of the scene that can be used by a search engine. Instead of analyzing a scene from a single perspective and storing only the corresponding meta-data, it is advantageous to capture relative object motion information as a descriptor that will ultimately support fast analysis of scenes on the fly from different perspectives, thereby enabling the ability to support a wider range of unexpected queries. For example, this can be very important in application areas such as security and surveillance, where it is not always possible to anticipate the queries.
  • One embodiment of the present invention is an apparatus and method for implementing object trajectory segmentation for an image sequence, thereby improving or offering other image processing functions such as context-based indexing of the input image sequence by using motion-based information. More specifically, block-based motion vectors are used to derive optical flow motion parameters, e.g., affine motion parameters. These optical flow motion parameters are employed to develop a prediction that is used to effect object trajectory segmentation for an image sequence.
  • optical flow motion parameters e.g., affine motion parameters.
  • optical flow (e.g., affine) object motion segmentation is initially performed for a pair of adjacent frames.
  • optical flow motion parameters between adjacent frames that describe the position of each point on a region at each time instant are made available to the present object trajectory segmenter.
  • the present invention is not limited by the method or model that is employed to provide the initial optical flow motion parameters between adjacent frames.
  • the object trajectory segmenter applies the optical flow motion parameters to form a new prediction or method for predicting the positions of all the points on an object over time within an interval.
  • the optical flow motion parameters are code fitted to form the new prediction.
  • the new prediction is then applied and the result is compared with an error metric.
  • the error metric measures the sum of deviations in distance at each point on the region at each time instant based on the new prediction compared to the original predictions.
  • the results from such comparison with the error metric will dictate the proper intervals (temporal boundaries) of the image sequence at which the motion parameters are valid for various key objects.
  • the present object trajectory segmenter obtains two sets of important information: the motion parameter values that accurately describe the object's motion and for which frames the parameters are valid.
  • the optical flow e.g., affine
  • the optical flow e.g., affine
  • motion trajectory such as direction, velocity and acceleration can be deduced for each key object over some frame interval, thereby providing an another aspect of motion information that can be exploited by query.
  • FIG. 1 depicts a block diagram of a content server of the present invention
  • FIG. 2 depicts a block diagram of a context-based indexer of the present invention
  • FIG. 3 depicts a flowchart of a method for implementing context-based indexing of an input image sequence by using motion-based information
  • FIG. 4 depicts a flowchart of a method for implementing optical flow (e.g., affine) object motion segmentation
  • FIG. 5 depicts a flowchart of a method for implementing optical flow (e.g., affine) trajectory segmentation
  • FIG. 6 illustrates code fitting of optical flow motion parameters to generate trajectory parameters
  • FIG. 7 illustrates a block diagram of an example as to which frames might be the temporal split points for an object that exists in a video sequence comprising of 20 frames.
  • FIG. 1 depicts a block diagram of a content server 100 of the present invention.
  • the content server 100 is implemented using a general purpose computer.
  • illustrative content server 100 comprises a processor (CPU) 140, a memory 120, e.g., random access memory (RAM), a context-based indexer 150, an optional encoder(s) 110 and various input/output devices 130, (e.g., a keyboard, a mouse, an audio recorder, a camera, a camcorder, a video monitor, any number of imaging devices or storage devices, including but not limited to, a tape drive, a floppy drive, a hard disk drive or a compact disk drive).
  • processor CPU
  • memory 120 e.g., random access memory (RAM)
  • context-based indexer 150 e.g., a context-based indexer 150
  • an optional encoder(s) 110 e.g., a keyboard, a mouse, an audio recorder, a camera, a camcorder, a video monitor, any number of imaging devices or storage devices, including but not limited to, a tape drive
  • the encoder(s) 110 and the content-based indexer 150 can be implemented jointly or separately. Namely, the encoder(s) 110 and the content- based indexer 150 can be physical devices that are coupled to the CPU 140 through a communication channel. Alternatively, the encoder (s) 110 and the content-based indexer 150 can be represented by one or more software applications (or even a combination of software and hardware, e.g., using application specific integrated circuits (ASIC)), where the software is loaded from a storage medium, (e.g., a magnetic or optical drive or diskette) and operated by the CPU in the memory 120 of the computer.
  • ASIC application specific integrated circuits
  • the encoder(s) 110 and the content-based indexer 150 (including associated data structures) of the present invention can be stored on a computer readable medium, e.g., RAM memory, magnetic or optical drive or diskette and the like.
  • various multimedia information are received on path 105 and stored within a storage device 130 of the content server 100.
  • the multimedia information may include, but is not limited to, various image sequences such as complete movies, movie clips or shots, advertising clips, music videos and the like.
  • the image sequences may or may not include audio streams or data streams, e.g. , closed captioning and the like. Due to the explosion of available multimedia content and their large size, the input information on path 105 may undergo a compression process that is illustrated as one or more optional encoders 110.
  • the encoders 110 may comprise video and audio encoders, e.g., MPEG-like encoders that are designed to reduce spatial and temporal redundancy. However, any number of compression schemes can be employed and the present invention is not so limited to any particular scheme.
  • the encoders 110 are optional since the input information may have already undergone various compression processes outside of the content server, where the input stream is already in a compressed format. In such implementation, the encoders 110 can be omitted.
  • the content-based indexer 150 is employed to analyze the input information and to provide an efficient index to the large quantity and often complex multimedia content that are stored on the storage device(s) 130.
  • the content-based indexer 150 of the present information is tasked to provide an indexing method and associated data structures that will allow an efficient method to categorize and then to allow retrieval of complex multimedia content quickly on path 195. More particularly, the present content-based indexer 150 employs motion information to allow more complex queries that employ
  • a query for an image sequence containing a blue background may generate a large number of positive query hits, thereby reducing the effectiveness of the query function.
  • the query can be modified for searching an image sequence containing a blue background with an object moving in the foreground at a high velocity to the left, then the response to the query may produce a highly focused set of positive responses, e.g., an image sequence having an aircraft moving rapidly across a blue sky background.
  • the content-based indexer 150 comprises an object motion segmenter 160 and an object trajectory segmenter 170.
  • FIG. 2 depicts a block diagram of a context-based indexer 150 of the present invention comprising an object motion segmenter 160 and an object trajectory segmenter 170.
  • the object motion segmenter 160 comprises a block-based motion estimator 210, an optical flow (e.g., affine) segmenter 212, a key object tracker 214, a key object splitter 216, and an optical flow (e.g., affine) segmenter 218.
  • an optical flow e.g., affine
  • the object trajectory segmenter 170 comprises a key object trajectory segmenter 220 and a sub-object trajectory segmenter 222.
  • the broad functions performed by these modules are briefly described with reference to FIG. 2. Detailed descriptions of these functions are provided below with reference to the flowcharts and other diagrams of FIGs. 3-6.
  • an image sequence is received into block-based motion estimator 210, where motion information, e.g., block-based motion vectors, are computed from the image sequence for each frame.
  • motion information e.g., block-based motion vectors
  • the content server 100 has an external encoder 110 or the input image sequence already contains motion information, i.e., where the motion vectors are encoded with the image sequence
  • block-based motion estimator 210 can be omitted.
  • the block based motion information can be extracted from the compressed bitstream itself or is provided by other modules of the content server 100, thereby relieving the object motion segmenter 160 from having to compute the motion vectors.
  • the optical flow (e.g., affine) segmenter 212 applies the motion vector information to generate "affine motion parameters" .
  • affine motion model is disclosed by J. Nieweglowski et al. in "A Novel Video Coding Scheme Based On Temporal Prediction Using Digital Image Warping", IEEE Trans. Consumer Electronics, Vol. 39, 3, pp. 141- 150, August, 1993, which is incorporated herein by reference.
  • the affine motion model constructs a prediction image or frame from a previous image by applying a geometric transformation known as "image warping". The transform specifies a spatial relationship between each point in the previous and prediction images.
  • motion compensation using block matching provides a good overall performance for translational motion.
  • the block-matching motion estimation is a poor performer when motion contains rotational or scaling components (e.g., zooming or rotating an image).
  • affine motion model (affine transformation) is defined by six parameters (ai to a ⁇ ) and is expressed as:
  • Key objects can be viewed as objects that are sufficiently significant that tracking of their motion is important for the purpose of indexing or other image processing functions.
  • key objects are identified in part based on their size, i.e., large objects are typically key objects, whereas small objects are not key objects.
  • a moving vehicle is typically a key object whereas a small moving insect in the background is not a key object.
  • the requirements for qualifying key objects are application specific, and are defined by the user of the present invention. Once key objects are defined, the motions of these key objects are then tracked by key object tracker 214.
  • a key object can be segmented into sub-objects and the motion information for these sub-objects can be tracked individually.
  • a key object of a human being can be segmented into six sub-objects comprising a head, a body and four limbs.
  • a query can now be crafted to search for "actions" that are relative to sub-objects within a key object, e.g., searching for an image sequence where a limb of a person is raised above the head of the person and so on.
  • the key objects information can be forwarded from the key object tracker 214 directly to an affine segmenter 218 for identification and segmentation of "sub-objects" for each key objects.
  • affine segmenter 218 is also tasked with generating affine motion parameters for the sub-objects.
  • the content-based indexer illustrates two affine segmenters 212 and 218, it should be understood that a single affine segmenter can be implemented to perform both levels of affine processing (i.e., key object and sub-object processing).
  • the motion information from key object tracker 214 is forwarded to a key object trajectory segmenter 220.
  • the motion information for each key object is forwarded to the key object trajectory segmenter 220, where motion trajectory information and intervals (frame intervals) are generated for each key object.
  • the motion information is summarized into "key object trajectory information", e.g., direction, velocity, acceleration and the like within some defined intervals (over a number of frames). This allows the motion information to be captured and stored in a format that allows for efficient motion-based indexing (or other image processing) of multimedia content.
  • FIG. 3 depicts a flowchart of a method 300 for implementing affine segmentation, thereby improving or offering other image processing functions such, as context-based indexing of an input image sequence by using motion-based information. More specifically, method 300 starts in step 305 and proceeds to step 310 where affine object motion segmentation is performed. Namely, key objects are identified within some intervals of the image sequence (also known as a "shot" having a number of frames of the input image sequence) and their motion information is extracted and tracked over those intervals. In step 310, affine motion parameters are generated for each identified key object.
  • affine motion parameters are generated for each identified key object.
  • step 320 the affine motion parameters generated for each identified key object for each adjacent pair of frames are processed over an interval of the image sequence to effect object trajectory segmentation. Namely, motion trajectory such as direction, velocity and acceleration can be deduced for each key object over some frame interval, thereby providing an another aspect of motion information that can be exploited by query. Method 300 then ends in step 325.
  • FIG. 4 depicts a flowchart of a method 310 for implementing affine object motion segmentation. Namely, method 310 is a more detailed description of step 310 of FIG. 3.
  • Method 310 starts in step 405 and proceeds to step 407, where method 310 generates affine motion parameters from block-based motion information. Namely, a random number of blocks are selected where their block-based motion vectors are employed to derive affine motion parameters as discussed below.
  • step 410 method 310 attempts to track one or more identified key objects from a previous frame, i.e., obtain the label of a key object from a previous frame. Namely, once key objects have been identified for a pair of adjacent frames as discussed in step 420 below, it may not be necessary to again apply the same detection step for the next frame in the image sequence. Namely, for a new frame, block based motion vectors can be employed to rapidly look backward to see whether the blocks point to a previously labeled key object. If so, such blocks will retain the same labeling as in the previous frame.
  • step 410 is skipped, as in the case where method 310 processes a new shot.
  • method 310 identifies the key objects within a pair of adjacent frames of the input image sequence.
  • the key objects are identified by determining whether a block is within the affine object flow.
  • method 310 may optionally merge identified key objects.
  • the identified key objects may be too small to be significant for image processing purposes, i.e., indexing in a particular application, e.g., a single bird can be merged with other birds within the frame to form a key object of a flock of birds.
  • method 310 may optionally identify sub-objects within key objects.
  • the identified key object may have significant motion information associated with its components (sub-objects) for indexing purposes in a particular application.
  • an identified key object comprising of a human being may comprise sub-objects, i.e., the person's limb, where the relative motion information of the sub-objects is important for indexing the shot.
  • method 310 queries if there are additional frames associated with the present "shot" . If the query is negatively answered, then method 310 ends to step 455. If the query is positively answered, then method 310 proceeds to step 407, where the affine motion parameters are generated for the next pair of adjacent frames.
  • FIG. 5 depicts a flowchart of a method 320 for implementing optical flow (e.g., affine) trajectory segmentation. More specifically, method 320 is a more detailed description of the steps 320 of FIG. 3 and the affine motion model is employed to describe the present invention.
  • optical flow e.g., affine
  • Method 320 starts in step 505 and proceeds to step 510 where affine motion parameters between adjacent frames to describe the position of each point on a region at each time instant are obtained as discussed above in FIG. 4.
  • various methods exist for determining the motion information for regions within a frame, e.g., various optical flow techniques.
  • the present invention is not limited by a particular method or model that is employed to provide the initial optical flow motion parameters between adjacent frames of an interval of the image sequence. Namely, as to the present trajectory segmenter, it is assumed that a "segmentation" or delineation between objects is previously computed and is known.
  • method 320 models at least one of the affine motion parameters for a particular or selected interval. For example, derive affine motion parameters between a subsampled set of frames in the chosen interval depending on the order of the fit, e.g., quadratic expressions require at least 2 data sets, etc. Specifically, decompose affine motion parameters into its components, namely, scale, rotation, shear, and translation. Assume different temporal models for each component depending on its nature. For example, the translation can be modeled using a polynomial, e.g., quadratic in time. The scale can be modeled to vary linearly over time. The rotation can be modeled using a constant angular velocity assumption.
  • a x and a y represent acceleration
  • v x and v y represent velocity
  • x 0 and y o represent the initial position.
  • the method begins with the initial estimate that the entire scene can be described by one set of motion parameters. In other words, the selected interval is the entire image sequence.
  • step 530 method 320 computes the trajectory parameters for the selected interval using code fitting, i.e., using the subsampled affine motions to obtain the coefficients for the chosen prediction model, represented by curve(s) 610ai-a6, as shown in FIG. 6. Namely, a new parametric expression or prediction is developed to predict the position of all the points on the object over time within a selected interval.
  • step 540 method 320 computes and sums the errors for each key object for at least one of the affine motion parameter. Namely, an error metric is chosen that measures the sum of deviations in distance at one or more points on the region at each time instant based on this new prediction as compared to the original positions.
  • step 550 method 320 queries whether the summed error is greater than a threshold "Tl" . If the query is positively answered, then method 320 proceeds to step 560, where the selected interval is split or divided into two intervals at the location of maximum frame error. If the query is negatively answered, then method 320 proceeds to step 570. Namely, the new trajectory parameter set is calculated. If the average error (averaged over the time interval for which the model is valid) exceeds a threshold Tl, then the sequence or selected interval is split into two temporal sections at the frame where there is a maximum error.
  • Tl threshold
  • step 570 method 320 queries whether there is a next interval. If the query is positively answered (as in the case when a split operation occurs in step 560), then method 320 returns to step 520, where steps 520-560 are repeated for each new interval, i.e., two new intervals are generated for each split operation. If the query is negatively answered, then method 320 proceeds to step 580. Namely, all intervals have been evaluated for splitting. Namely, method 320 continues to calculate the motion parameters of each interval as well as the error for that time interval, and continues to perform a temporal split if the average error exceeds the threshold. The iteration stops either when the time interval is smaller than 3 frames or when the error is below the threshold.
  • Sections or segments 0 and 2 are considered not "valid" because these sections are required to be split.
  • Sections 1, 3 and 4 are valid since no splitting is required. Only the motion parameters of the valid sections are stored, and passed on to the Merge subroutine. It should be noted that the start and end frames of Sections 1, 3, and 4 are contiguous. As described below, it is plausible that Sections 1 and 3 can be combined into a single section whose motion parameters fall within the error threshold.
  • step 580 method 320 evaluates each pair of adjacent intervals for potential merging. Specifically, adjacent intervals are merged, if a joint prediction model for two adjacent intervals results in a normalized error below a threshold "T2". This step is repeated for all intervals recursively until any further merging increases the normalized error in the merged interval above the threshold, T2.
  • T2 a threshold
  • a merge operation is applied to all the "valid" time intervals. The merge operation looks at successive temporal segments or intervals, and calculates a new set of trajectory parameters for the two selected segments, and decides to merge the two segments only if the error (of the parametric motion model of the merged segments) falls below a threshold.
  • Method 320 ends in step 590 when all possible merge operations have been performed.
  • method 320 of the present invention provides an effective way of describing the motion parameters of an object within a sequence, and detecting the temporal boundaries for which to update the motion parameters, given a motion model. These resulting descriptors can then be exploited by image processing functions, e.g., applied to a video sequence for indexing or searching tool, to detect objects or events consisting of object interactions.
  • An example illustrating the importance of the split and merge operations is as follows: Suppose a sequence starts with frames 0-20, and it has been determined that the best split point is at frame 5. After a split operation is performed at the specified location, two segments or intervals, frames 0-5 and 6-20 remain.
  • frames 0-5 need not be split, and the best split point for the interval from frames 6-20 occurs at frame 8, then after another split operation, three segments: frames 0-5, 6-8, and 9-20 remain. Once all segments no longer need to be split, a successive check is performed to determine if adjacent segments can be merged. Thus, in the above examples, a check is performed to determine if the first two segments (frames 0-5 and frames 6-8) could be merged into one segment forming frames 0-8, thus if the merge occurs a total of two segments, frames 0-8 and frames 9-20 are obtained. It should be noted that in the first pass of the split operation, the best split point was at frame 5 and not at frame 8.
  • AvgPos average spatial position of Object at each frame
  • NumSections Me ⁇ ge(SectionParameters , NumSections , AvgPos, Threshold); ⁇
  • NumSections Split(SectionParameters , NumSections, AvgPos, Threshold);
  • NumSections Split(SectionParameters, NumSections, AvgPos, Threshold); ⁇ ⁇ ⁇
  • the above method illustrates a general approach that allows for modeling of the object's motion via any parametric model.
  • error metric can be computed to determine whether or not the object's true motion is well described by the parametric model.
  • split and merge technique can be used to find the boundary points of when the model's parameters need to be updated.
  • the present invention is described using an affine motion model that describes an object's translational, rotational, shear, and zoom characteristics, it is not so limited.
  • v x and v y represent the velocity
  • C and F represent the translational motion components
  • A, B, C, and F describe the rotational, shear, and zoom components:
  • the above-discussed thresholds are application specific as each application can have different levels of tolerance to errors in trajectory.
  • the threshold that can be used for the above quadratic case is as follows: If (sum of deviation in distance over all points in the interval)/(Number of points in the interval) > 0.9, then split. Similarly, if (sum of deviation in distance over all points in the interval)/(Number of points in the interval) ⁇ 0.9, then merge.
  • the present object motion segmentation and object motion trajectory segmentation are described above for improving the indexing and retrieval of image sequences, it should be understood that the present invention can be employed to improve or provide other image processing functions.
  • the present invention can be employed in image processing functions, e.g., the synthesis of content from object trajectory (for quick preview) given the initial texture.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Library & Information Science (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

Cette invention concerne un appareil (170) et un procédé permettant de réaliser la segmentation de la trajectoire d'un objet pour une séquence d'images. Spécifiquement, on utilise des vecteurs de mouvement à base de blocs correspondant à une paire d'images adjacentes pour déterminer des paramètres de mouvement de flux optique, des paramètres de mouvement affine par exemple. Les segments de trajectoire de l'objet appliquent les paramètres de mouvement de flux optique de façon à constituer une nouvelle prédiction ou un nouveau procédé de prédiction des positions de chacun des points par lesquels passe un objet durant un certain intervalle de temps. On applique alors la nouvelle prédiction et on compare le résultat avec un système de mesure d'erreurs. Les résultats obtenus par une comparaison de ce type permettent de définir des intervalles appropriés (limites dans le temps) de la séquence d'images, durant lesquels les paramètres de mouvement sont applicables à différents objets clé.
PCT/US2000/002364 1999-01-28 2000-01-28 Appareil et procede de description des parametres de mouvement d'un objet dans une sequence d'images WO2000045339A1 (fr)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US11764999P 1999-01-28 1999-01-28
US60/117,649 1999-01-28
US48962600A 2000-01-24 2000-01-24
US09/489,626 2000-01-24

Publications (1)

Publication Number Publication Date
WO2000045339A1 true WO2000045339A1 (fr) 2000-08-03

Family

ID=26815488

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2000/002364 WO2000045339A1 (fr) 1999-01-28 2000-01-28 Appareil et procede de description des parametres de mouvement d'un objet dans une sequence d'images

Country Status (1)

Country Link
WO (1) WO2000045339A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005006762A3 (fr) * 2003-07-02 2005-02-10 Queen Mary & Westfield College Procede d'estimation de flux optique

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0805405A2 (fr) * 1996-02-05 1997-11-05 Texas Instruments Incorporated Détection d'événements de mouvement pour l'indexation de vidéos

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0805405A2 (fr) * 1996-02-05 1997-11-05 Texas Instruments Incorporated Détection d'événements de mouvement pour l'indexation de vidéos

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
NIEWEGLOWSKI J ET AL: "A NOVEL VIDEO CODING SCHEME BASED ON TEMPORAL PREDICTION USING DIGITAL IMAGE WARPING", IEEE TRANSACTIONS ON CONSUMER ELECTRONICS,US,IEEE INC. NEW YORK, vol. 39, no. 3, 1 August 1993 (1993-08-01), pages 141 - 150, XP000396273, ISSN: 0098-3063 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005006762A3 (fr) * 2003-07-02 2005-02-10 Queen Mary & Westfield College Procede d'estimation de flux optique
US7822231B2 (en) 2003-07-02 2010-10-26 Queen Mary And Westfield College, University Of London Optical flow estimation method

Similar Documents

Publication Publication Date Title
US6366701B1 (en) Apparatus and method for describing the motion parameters of an object in an image sequence
US6643387B1 (en) Apparatus and method for context-based indexing and retrieval of image sequences
Gunsel et al. Temporal video segmentation using unsupervised clustering and semantic object tracking
US7986810B2 (en) Mesh based frame processing and applications
Shen et al. A fast algorithm for video parsing using MPEG compressed sequences
US7447337B2 (en) Video content understanding through real time video motion analysis
Hariharakrishnan et al. Fast object tracking using adaptive block matching
JP3719933B2 (ja) 階層的ディジタル動画要約及び閲覧方法、並びにその装置
Zhong et al. Video object model and segmentation for content-based video indexing
Zhong et al. Spatio-temporal video search using the object based video representation
US7142602B2 (en) Method for segmenting 3D objects from compressed videos
US20080043848A1 (en) Video/audio signal processing method and video/audio signal processing apparatus
Nasreen et al. Key frame extraction from videos-A survey
Schonfeld et al. Vortex: video retrieval and tracking from compressed multimedia databases—multiple object tracking from mpeg-2 bit stream
Babu et al. Compressed domain video retrieval using object and global motion descriptors
WO2000045339A1 (fr) Appareil et procede de description des parametres de mouvement d'un objet dans une sequence d'images
KR100361939B1 (ko) 객체 움직임을 이용한 ｍｐｅｇ 비디오 시퀀스의 데이터 베이스 구축 및 검색 방법과 그 기록 매체
Wang et al. Real time motion analysis toward semantic understanding of video content
Bocheck et al. A content based video traffic model using camera operations
Yongsheng et al. A Survey on Content based video retrival
Kuo et al. Efficient algorithm for scene change detection and camera motion characterization using the approach of heterogeneous video transcoding on MPEG compressed videos
Padmakala et al. A novel video object segmentation approach for noisy video sequences towards effective video retrieval
Venkateswaran et al. DWT based hierarchical video segmentation
Ashwini et al. Analysis of Global Motion Compensation and Polar Vector Median for Object Tracking Using ST-MRF in Video Surveillance
Schweitzer Computing content-plots for video

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): BR CA CN JP KR

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE

121 Ep: the epo has been informed by wipo that ep was designated in this application
DFPE Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101)
122 Ep: pct application non-entry in european phase
点击 这是indexloc提供的php浏览器服务，不要输入任何密码和下载