Algorithm Research And Implementation Of Person Re-identification For Video Surveillance Based On Deep Learning

With the development of artificial intelligence technology and the implementation of smart city policy,person re-identification technology has become a hot research topic in the field of artificial intelligence.Person re-identification plays an important role in public safety with significant potential but also great challenges.In practical applications,on the one hand,person re-identification algorithms are difficult and slow to recognize due to factors such as occlusion and gesture changes;on the other hand,the current research on person re-identification technology focuses on the stage of separate person identification algorithms,and more detailed research is needed from the algorithm to the actual implementation.To address the above issues,this paper investigates the algorithms design and system implementation of video surveillance person re-identification using deep learning technique.The specific contents are as follows:(1)A person re-identification algorithm based on the combination of attention mechanism and gesture recognition was designed for the problems of occlusion and gesture variability.In the designed network,spatial and channel global attention network and gesture recognition network were used to acquire global features,joint position heat maps and corresponding probability values,and also proposed to obtain local features by using the computation of position heat map.After fusion with global features,multi-task and multi-loss were used to jointly supervise the optimization of network.The Rank-1and m AP metrics reached 85.1% and 75.6% on the Market1501 dataset.The results show that the proposed algorithm has the ability to resist gesture change,occlusion and background,and also has high recognition capability and recognition accuracy.(2)To overcome the problem of low recognition accuracy due to incomplete extraction of person features by traditional methods using only convolutional neural network architecture,a sequence Transformer video person re-identification algorithm based on a priori knowledge was proposed.The features of the middle layer of the Res Net50 network was used as a priori features of the Transformer architecture to improve the robustness of the model.On the MARS dataset Rank-1 and m AP reached86.8% and 80.3% respectively,an increase of 3.8% and 3.3% over the benchmark.The results show that the designed algorithm is fast and accurate in recognition,and can also be used as a base network for feature extraction.(3)In order to verify the actual recognition effectiveness of the proposed algorithms,the performance of the algorithms was tested in real scenes.After detecting and cropping person images from surveillance video using the YOLOv5 model,the two algorithms were tested in indoor and outdoor scenes respectively to demonstrate the effectiveness of the proposed algorithms.At the same time,a strategy using distance thresholds was proposed in order to cope with the need for real-time person re-identification in real-life situations.(4)A person re-identification system was designed and implemented.Two proposed person re-identification algorithms were used to implement person detection,rankingbased and real-time person re-identification functions end-to-end by combining them with person detection algorithm.Finally,the person re-identification system function was tested in an actual surveillance video.It is proved that the system can accomplish the expected functions and has fine theoretical research and practical application value.There are 50 figures,9 tables and 83 references.