Research On Person Re-identification Method Based On Monitoring

In recent years,with the continuous rise of deep learning technology represented by convolutional neural network in the field of computer vision,and its excellent performance,person re-identification(re-id),a new computer vision task,has a new development.Person re-id task can help police to find lost children,track suspects and reduce police force consumption with the help of surveillance cameras all over the country.However,the main research in pedestrian recognition industry is based on the existing open database,so the research object of various algorithms is not the surveillance cameras all over the streets of our country,and the mainstream pedestrian recognition algorithms in the industry can't effectively process the videos and pictures captured by the surveillance cameras.In this paper,we studied the person re-id method based on deep learning,and the original method is modified to make it more effective to study the images captured by the surveillance camera.The main contributions of this paper are as follows:In view of the fact that the characters in the real surveillance camera are all the depression view angle,while the characters in the open data set used by the academic research institute are the horizontal view angle,this paper improves on the two methods of partial convolution network and global average pool layer,and proposes a pedestrian recognition method based on "region of interest".This method improves the "horizontal stratification" method in the partial convolution network method,proposes the nonparametric spatial attention layer,designs a more robust pedestrian recognition model,and focuses on solving the problem of pedestrian recognition under the depression view angle in the actual monitoring image.In view of the problem that the pedestrian object captured by the actual monitoring camera often appears incomplete and occluded,this paper proposes a pedestrian recognition method based on the edge contour information of human body.By measuring and learning the pedestrian image,the binary mask of pedestrian contour information and the background of pedestrian image,the model can better deal with the complex background in the actual monitoring Common problems.