Robust Tracking Of Moving Object Based On Feature Combination And Multiple Kernel Learning

With the development of video analysis technology, multimedia databases and artificial intelligence technology, intelligent video surveillance has been gradually applied in security applications market. Intelligent technology extracted from the original video a lot of useful information timely and automatically, to complete the video saving and retrieving transfer, can also drive other data, trigger other actions, it can easily complete task which is difficult to do by human. Therefore, the study of intelligent video surveillance system and improve the performance of the technology has important significance.Tracking of moving object in video surveillance is an important component of the system, and other studies of computer vision have an important role in promoting. In this paper, we have a thorough study on video tracking. In the current research in this field, we research the theory and methods of tracking of moving objects in intelligent video surveillance.We adopt a method which present the object using spatial pyramid with SIFT feature, it includes orderly geometric correlations among features. Then incremental PCA as a guider, an updating mechanism---Online Adaboost classifier is adopted to deal with pose and appearance changes of object. Most experiments demonstrate the robusty of our method in object tracking successfully. Our approach is more robust in handing occlusions and some challenge situations.By combining spatial pyramid of Histogram of Orientation Gradients (PHOG) and Scale Invariant Feature Transform (PSIFT) as the "strong feature", we present a new tracking method capable of handling occlusions. By simultaneously adopting PHOG and PSIFT, shape information and appearance information are fused. Besides, we use an Online Adaboost classifier and update the classifier through selecting the most distinctive samples to retrain it.We also present a new method for object tracking based on multiple kernels learning (MKL). MKL is used to learn an optimal combination of kernels, each type of which captures a different feature. Our features include the color information and spatial pyramid histogram. On-line updating MKL classifier is adopted, where useful tracking objects are automatically selected as support vectors. The algorithm handle target appearance variation, and makes better usage of history information.