Human Action Recognition In Video Sequence

Human action recognition in video sequence is a hot topic in the field of computer vision.Its main task generally includes detecting the moving object from the video sequence,extracting the action characteristic,representing the movement information and carrying on the movement classification.It involves the content of pattern recognition,digital image processing,mathematics,computer vision.It has important theoretical value and application value.This paper mainly includes three parts: action target detection,feature extraction,human action classification.The first part mainly introduces several common methods of moving target detection,and compares the advantages and disadvantages.We analyze the background subtraction method used in the experiment and the algorithm of mixed Gaussian background model in detail.The second part mainly includes action feature extraction and feature description.This paper first introduces two commonly used optical flow algorithms,and analyzes their advantages and disadvantages.Based on the shortcomings of Lucas-Kanade optical flow algorithm introduces the improved algorithm of Lucas-Kanade optical flow,which is Pyramid optical flow method.Then,the paper introduces to two commonly used descriptorsin detail: directional gradient histogram and optical flow direction histogram.On the basis of this,we analyze the descriptor of the weighted optical flow direction histogram,and use it to describe the action flow characteristics.The third part uses the support vector machine to train and classify the extracted optical flow characteristics.Of all,this chapter introduces the classification principle,characteristics and application of support vector machine.Next,Then the simulation experiment is carried out on the Weizmann data set and analyzed the experimental results.