• David Steketee (21962175)
• Joshua Pollard (21966798)
• Thomas Hill Almeida (21963144)

The aim of this project is to build an object recognition system in MATLAB to recognize and locate reference images in a cluttered scene.

Object detection and location is done using SURF features to be able to detect objects present in these images with reasonable accuracy even with significant occlusion and clutter in the scene images. The example scenes and reference images in this repository are classified among three difficulty levels; easy, difficult and very difficult.

Easy scene images are those with no occlusion of target objects. Difficult scenes are those some occlusion of target objects. Very difficult images typically have significant occlusion of target objects and clutter in the image.

There are a number of folders that are included as a part of this repository.

• quick_run_pics is a folder that contains a small number of reference and scene images on a black background that can be quickly loaded and run
• full_set_pics is a folder that contains a large number of reference and scene images on a white background that take longer to process and run
• readme is a folder that contains pictures used in this README

This project requires MATLAB in order to run, and requires that this git repository is in the PATH so that it can access the scripts needed to run.

It also requires the following MATLAB add-on is installed:

• Computer Vision Toolbox

When the GUI is first loaded, it will look like this:

There are a number of basic usage instructions that can be seen in the bottom left of the interface.

To get started:

1. Select Process reference images on the right and select a directory of reference images
   • The directory chosen must be a directory containing directories named as the name of the reference object, with each directory having some number of reference images in them
   • Once the images have finished loading, the status box on the bottom left will display the message "training images loaded"
2. Select Save current model on the right to save the trained model to a .mat file
   • Previously saved structures can be loaded using the Load pre-processed model button on the right to quickly reload a trained model.
3. Select Load Scene on the top left and select a scene image to analyse.
4. Toggle the checkboxes on the left to select if lines should be connected between features in the scene or if object outlines should be drawn.
5. Select Isolate objects to begin the search.
   • As reference images are found, they will be appended on the right hand side of the scene image.
   • A list of the found objects will be printed to the status box on the bottom left.