Research Of Person Re-identification In Video Based On Deep Learning

With the development of science and technology in the field of security monitoring and the country's emphasis on the field of security monitoring,China has become one of the safest countries in the world.The non-overlapping and multi-view camera pedestrian re-identification is the task of solving the same pedestrians under cameras with different views.This task is of great significance to the administrative agencies and judicial organs in maintaining social public safety.Due to the limited resolution of the camera itself,it will result in a low pedestrian resolution observed by the camera,and at the same time there is a difference between the perspective,the light and the captured pedestrian attitude between cameras of different visual areas.These factors will identify the same pedestrian for the algorithm to match.The process interferes,making pedestrian recognition in a non-overlapping view is still facing great challenges.In order to solve these problems,thesis conducts in-depth research on pedestrian-recognition based on video-based non-overlapping views.The main research works and innovations of the thesis are as follows:1.Based on the video re-identification of pedestrians under non-overlapping multi-view cameras,this thesis proposes a video pedestrian recognition method based on fusion generation against network and pose estimation.This method increases the number of samples and adds the feature information of pedestrians joints by generating anti-network networks to generate video prediction frame sequences to improve the efficiency of the model.Experimental results show that the improved method proposed in this thesis can effectively improve the recognition rate of public data sets.2.The differences in the motion cycle and pace between pedestrians in videos taken by different cameras influence the convolutional neural network to extract motion information and interfere with the expression of motion information.This thesis designs a video pedestrian recognition method based on multi-scale frame rate.Firstly,an adaptive separable convolution interpolation frame algorithm is used to insert frames into the dataset.The generated image is clear and the pedestrian motion trajectory is maintained.Then various frame rates are used to train the time-space attention network at random,and the frame rate is adapted.The changes have also improved the experiment.