Research On Human Behavior Recognition Method Based On Improved Dense Trajectory

Human action recognition is a hot research topic in the field of machine vision and artificial intelligence.It has wide application prospects in the fields of video surveillance,content-based video retrieval,human-computer interaction,intelligent transportation and so on.Due to the non-rigid motion,the complexity of the background,the mutual occlusion between people and the camera movement,human action recognition is a challenging topic.In recent years,researchers have made great progress in related fields,especially the human action recognition algorithm based on dense trajectory has achieved better recognition results than previous algorithms.An improved dense trajectory-based approach is proposed to recognize human action.Firstly,dense optical flow is utilized to track the scale invariant feature transform key-points at multiple spatial scales.In real scene video,which camera motion is intense,massive trajectories exist in the background.To eliminate the influence of camera motions,we via the analysis of saliency and the consistence indirection to improve the robustness of trajectory.Secondly,the space time pipeline is built and then is divided into space-time grid.The histogram of oriented,gradient histogram of optical flow and motion boundary histogram are extracted based on the space-time grid.In order to reduce the computational complexity,the principal component analysis(PCA)is used to reduce the dimension of each feature.Finally the fisher vector model is utilized to compute one vector for each descriptor separately,and the linear support vector machine is employed for classification.Through the improvement of dense optical flow,a complete action recognition method is formed.Experimental results on KTH and YouTube datasets demonstrate that the proposed algorithm can effectively recognize human actions.