Research On Human Detection And Tracking Technology Based On Surveillance Video

Intelligent monitoring system is a system which has the function of active monitoring and early warning.Moving target detection and tracking is an important part of intelligent monitoring system.Based on the requirements of monitoring system,moving target detection algorithm and tracking algorithm of surveillance video are studied in this paper.Firstly,the maximum and minimum luminance component value is substituted by the mean of the first k maximum and minimum luminance component values in the pixel,and the idea of adaptive Kalman filter is introduced,which improves the traditional codebook model and reduces its sensitivity to light change.The experimental results show that the improved Codebook model can significantly reduce the noise in the background.On this basis,a moving object detection algorithm based on codebook model and edge detection is proposed.Firstly,an improved Codebook model is used to detect the video,and then the edge detection is used to detect the edge of the current frame image and the edge of the mean background,and finally the detected moving target is manipulated.The experimental results show that the detected moving target is complete and accurate and the edge is clear.At the same time,the algorithm improves the foreground extraction rate by 24% comparing with the GMM algorithm,the improved Code model algorithm is 5.3%,and the processing can reach 57 frames/sec,which can meet the real time requirement.After that,aiming at the shortcoming of the depth feature in the rotation adaptation,this paper presents a new fusion algorithm,which integrates the rotation invariant feature of the LBP texture and the correlation filtering of the convolution neural network.After constructing a convolution neural network and training it off-line,the texture feature is combined with the layered convolution feature by the weighted fusion method,and the scaling adaptive and updating strategy are adopted.The experimental results show that the algorithm of this paper can improve the tracking accuracy by 11.8% and track success rate by 2%,and has better robustness to the rotational deformation and scale change of the target than the same HCF algorithm using convolution feature.Finally,a monitoring interface is designed to verify the detection and tracking algorithm in this paper,the experiment proves that the algorithm can detect and track the human in the actual monitoring scene.