Cognitive Object Recognition System - It is all in the brain!

Automated object recognition is an important and challenging problem. The technology is crucial for a number of Army applications including video surveillance, Automatic Target Recognition and Simultaneous Localization and Mapping. While feature-based or template matching-based classification algorithms are used in certain Army applications, it has been observed that their classification accuracy depends on the quality of the training dataset. Often the lack of a representative set of training images severely affects performance. Moreover, the algorithms generally perform poorly when the object is occluded. Each of these methods addresses certain aspects of recognition while ignoring others. In phase I of this project, an innovative psychologically-inspired approach in which Geon-theory and Feature-based recognition algorithms are integrated together to provide a comprehensive solution to robust object recognition was successfully explored. In phase II UC will enhance and mature the first-of-its-kind generic object recognition system to recognize objects of strategic interest to the Army. Object Search/Top-down Saliency will be incorporated into the system thereby allowing active user interaction through a user interface. Landmarking functionality will be developed to enhance localization capability of current SLAM product. Platform independent Cognitive Object Recognition System and a Context identification/image processing robotics toolbox will be implemented.