Human Motion Detection And Posture Recognition Algorithm Research On Based On Computer Vision

Human motion detection and posture recognition directly use the computer to analyze and process the visual information of the video,detect the human motion and posture,describe and express the information and purpose that the posture would convey.With the development of computer technology and information technology,the society has higher requirements for video analysis,and hopes to get the human posture information directly through the analysis of the video by computer.Because of its deeper understanding of the human visual system,human posture research is one of the key technologies to be realized in the field of computer vision.This paper takes the human motion of the indoor environment as the research object,and uses the still cameras to capture videos with moving human objects.First of all,based on the image preprocessing,the method of moving target detection is studied deeply.Considering the shortcomings such as trailing smear and lower performance of the traditional GMM in the process of moving object detection,this paper proposed an improved Gaussian mixture model algorithm fusing Wronskian function and frames difference method.After the process of Gaussian mixture model,the spatial domain correlation between neighboring pixels are judged by the value of the Wronskian matrix determinant.So the improved algorithm increases the update condition of the model parameters and improves the updating mechanism of the model parameters.At the same time,using the sensitivity of frames difference method to detect moving target contour,it applied Boolean OR operation on the results of improved GMM and frames difference method to get the preferable moving object.Secondly,in the research of human posture recognition method,this paper proposed a human posture recognition algorithm based on multi-feature fusion and image similarity.The double threshold segmentation algorithm is used to extract the foreground object,and extract and construct the human posture model from the normalized foreground object.The human posture model includes the local contour pole distance feature,the local contour pole angle feature and the eccentricity.The template matching method based on SVM is used to study and train the multi-feature of human posture.By calculating the similarity of two adjacent images,the classification results of SVM are classified secondly to increase the degree of correlation between the adjacent frames in the recognition process.Finally,this paper established a test system to test the standard video libraries and self-built video libraries.The experimental results show that the improved human motion detection algorithm can effectively suppress the phenomenon of the trailing smear and enhance the detection performance compared with the traditional GMM and frames subtraction algorithm.And the improved human posture recognition algorithm can effectively identify the human posture and has lower computational complexity,which overcomes the limitation and uncertainty of the result that only takes the single frame classification as the final result,increases the degree of correlation between the video adjacent frames in the recognition process,and improve the accuracy of posture recognition while guaranteeing the real-time performance.