Experimenting WNN support in object tracking systems

Object tracking is a challenging problem in many computer vision applications, which go from robotics to surveillance systems. When applied to real world conditions, tracking methods found in the literature compete in solving some inherent difficulties of object segmentation and movement prediction, such as camouflage, occlusions, dynamic background, brightness, color and shape changes. To address some of these issues, we propose a general framework for object tracking by exploiting well-known segmentation techniques and a weightless neural network based prediction algorithm. The considered neural computing model is DRASiW, that we, here, extended with reinforcing and forgetting mechanisms. This model has the property of being noise tolerant and capable of learning step-by-step the new appearance of the moving object, by updating the learned object shape through the evolution of its internal representation (called “mental” image). The proposed object tracking framework has been evaluated on different benchmark videos. Experimental results show the viability and the benefits of the proposed DRASiW-based object tracking framework in the chosen case studies in comparison with three state-of-the-art methods. In addition, results provide useful insights about which combination of DRASiW-based operational modes and segmentation techniques improves the performance in the considered cases.