文摘
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.