ITEE Innovation Expo 2008 : Project Details

Biologically Inspired Object Recognition, Classification and Tracking Toward Real-Time Applications

The goal of enabling a computer to visually interpret the real world that surrounds it has long been a goal of computer science. Despite all of the advances in computational power and understanding of visual function in humans and closely related primates, the ability of a computer algorithm to recognise and classify objects at speeds and accuracies approaching that of the human brain has remained elusive.

This thesis examines three main research topics and the linkages that can be made between them. The first is biological vision systems, focussing on humans and other primates, used as a basis for comparison to novel computer vision implementations. The second is the computer implementation striving to mimic or surpass those biological systems in allowing computers to identify and relate to real world objects – object recognition and image classification. Lastly, computer algorithms and theories for tracking objects in the real world across time-series images or video are examined. Attempting to model a more holistic system for a computer vision system, methods of linking an accurate, albeit slow, object recognition and classification model with an algorithm for tracking through an image stream in real time are examined and implemented.

While this linked model still suffers from the slow processing speed of the object classification algorithm, the addition of the tracking algorithm removes the need for reclassification for objects already indentified in previous frames in an image stream. Further refinements to the algorithms and overall model affecting accuracy, speed and overall performance are discussed.