In this paper, we propose an automatic tracking recovery tool which improves the performance of any tracking algorithm each time the results are not acceptable. For the recovery, we include an object identification task, implemented through an adaptable neural network structure, which classifies image regions as objects. The neural network structure is automatically modified whenever environmental changes occur to improve object classification in very complex visual environments like the examined one. The architecture is enhanced by a decision mechanism which permits verification of the time instances in which track-ing recovery should take place. Experimental results on a set of different video sequences that present complex visual phenomena reveal the efficiency of the proposed scheme in proving tracking in very difficult visual content conditions, abstract environment.
展开▼
