Multi-feature Based Unspecific Video Target Recognition Algorithm

Target surveillance and identification algorithms have always been a vital part in the research area of computer vision, which are mainly applied in numerous automatic devices, such as intelligent surveillance systems, robots and bio-medical instruments. Currently a variety of recognition methods based on solitary feature have been well proposed, while all these techniques could properly achieve their goals under certain conditions. The main content of this paper is, by investigating and analyzing such existed methods, to find an approach synthesizing different image features to implement target recognition, thereby improving its adaptability and robustness.Firstly, this paper analyzes some commonly used motion detection methods, and summarizes their respective features. Then some study concerning a few typical image features and respective application in target recognition is presented. After that, we discuss a specialized research in active contour model (snake). And finally, all these studies form the structure of our proposal, including motion detection, edge extraction and feature identification.Experiments indicate that the proposed method can efficiently follow the alternation of target features, therefore it could correctly accomplishes the task for tracking deformable targets, which couldn’t be solved by traditional sole feature based recognition methods. It also serves as the prototype for vertical research concerning intelligent recognition systems.