Research On Video Classification Technique Based On Lie Group And Dynamic Texture

Automatic video classification is a very important but challenging problem in theresearch of computer vision, and more and more scholar take research in it. The purposeis to classify the video by their category automatically by computer to replace manuallyclassification. However, existing video modeling method neglect the topology structureof the model, and is inaccurate on their measure. Therefore, the Lie group manifoldtheory is introduced to analysis the dynamic texture model.Recently, modeling method based on dynamic texture was introduced in the study ofvideo classification, and attract extensively attention. As it’s perfect characteristics atcapture the apperance and motion information of video, dynamic texture modelingmethod achieve highly effect. Though existing metric measure of dynamic textureneglect the topology structure of it, and is inaccurate. A reasonable metric measureshould take consideration in topology structure of the model and accurately classify thedynamic texture on such a accrate manifold. There is a map between dynamic texture andLie group manifold. This map and measure dynamic textures on Lie group manifold wasproposed. Then a kernel was constructed through the map between dynamic texture andLie group, and combine it with the support vector mechine method to improve the effectsof dynamic texuture classification. To verify the effectiveness of our method, experimentwas executed on traffic classification dataset. Experiment shows that the proposedmethod achive higher accuracy that other similar method.