Human Behavior Recognition Based On Vision

The field of computer vision is rapid development In the decades,many research results have already permeated all aspects of our life.The application of human behavior recognition has been widely spread,for example,intelligent monitoring,human interaction and retrieval are all products derived from the principles of human behavior.Because it is closely related to people's life,it has been favored by many scholars and has shown great application and economic value.This has also led to a continuous improvement in the social demand for human behavior,even though the technology has become one of the most concerned reasons.At the same time,it is also a major challenge for us.In this paper,the research on human behavior recognition technology has the following contents:First,human behavior target detection: mainly through background subtraction,first use the average value of the first frames of the video sequence to construct the background frame,and target prospects are segmented from the background by making difference between the video frames and the background frames.We use an improved algorithm of the global threshold of another method,Otsu,to extract the target foreground.The main principle is to divide the image's pixel value by two values by selecting the appropriate threshold method,and divide the image background and foreground region.Compare the effect of the two methods of target detection,choose the better method to carry out the next step.Two,human behavior feature extraction: in feature extraction,we mainly extract the shape features of the target behavior.(Histogram of Gradient,HOG)and(Fourier Descriptor,FD)in order to get various features to improve recognition rate,this paper chooses to extract SIFT characteristics of target behavior,and finally uses optical flow method to extract feature trajectories.The above features are fused to get the final behavior.The effectiveness of the method is verified by experiments.Three,human behavior classification: This paper uses the support vector machine SVM classifier to classify the human behavior.In the MATLAB simulation environment,four groups of comparative experiments were done by using KTH database and Weizmann database.The human behavior characteristics were extracted by HOG,HOG+FD,HOG+PCA+SIFT and the algorithm respectively.The effectiveness and robustness of the proposed algorithm are proved by experiments.