Feedback-based parameterized strategies for improving performance of video surveillance understanding frameworks

Abstract

One of the most ambitious objectives for the Computer Vision research community is to achieve for machines similar capacities to the human’s visual and cognitive system, and thus provide a trustworthy description of what is happening in the scene under surveillance. Most of hierarchic and intelligent video-based understanding frameworks proposed so far allow the development of systems with necessary perception, interpretation and learning capabilities to extract knowledge from a broad set of scenarios, having in common the one-way sequential structure of the functional processing units that compose the system. However, only in a limited number of works, once visual evidence is achieved, feedback is provided within the system to improve system’s performance in any sense. With this motivation, a methodology for introducing feedback in perceptual systems is proposed. Experimental results demonstrate how different parameterized strategies let the system overcome limitations mainly due to sudden changes in the environmental conditions.