Study On Clothing Change Person Re-identification Based On Pose Normalization And Knowledge Driven

Pedestrian re-identification is an extension of face recognition technology and one of the core applications in the field of intelligent monitoring.Pedestrian re-identification is designed to identify target pedestrians in different time and space from a distributed multi-camera system with non-overlapping views.Target pedestrians are susceptible to lighting,posture changes,and perspective changes in real scenes.In addition,pedestrians often change their clothing features due to external factors,how to ensure clothing consistency in long-term scenes has become an important challenge for pedestrian re-identification.In order to solve the above problems,this paper makes an in-depth analysis of existing pedestrian re-identification methods,proposes a pose normalization method based on style erasing and dual-channel fusion,and a clothing change re-identification method based on knowledge-driven and topology inference.Based on the above research,a pedestrian change recognition system is designed and implemented.The research work in this paper is as follows:(1)A posture normalization method based on style erasing and dual-channel fusion is proposed to improve the robustness and accuracy of pedestrian re-identification in the context of posture change.Firstly,erase the style in the image by using instance normalization,and then a content attention mechanism is proposed to restore the pedestrian’s discriminatory features.Secondly,a dual-channel fusion posture normalization module is designed,which uses the posture encoder to extract pedestrian posture information to assist the re-identification,and reduces the re-identification error caused by the misaligned features in the pedestrian matching process.Experiments show that the method can effectively improve the re-identification of pedestrians with different postures.(2)A re-identification method for clothing change based on knowledge-driven and topology inference is proposed to improve the recognition accuracy of pedestrians in changing clothes scenarios.Firstly,the semantic relationship between external factors and camera topology is captured by constructing a knowledge graph,and the acquired knowledge is embedded into the spatio-temporal graph convolution to improve the accuracy of topology inference.Secondly,the camera logical topology information is used to extract clothing information with strong correlation,and the auxiliary information is used to predict the clothing features of the target pedestrians in a specific period in order to reduce the negative effects of clothing changes.The results obtained after comparison experiments and ablation experiments show that this method can effectively improve the recognition of pedestrians in the case of clothing change.(3)Based on the above research,design and implement a pedestrian re-identification system for changing clothes scenarios.The system mainly includes information upload module,style normalization module,camera logic topology inference module,clothing change pedestrian re-identification module,and information storage module.Experiments show that this system can effectively achieve rapid retrieval of target pedestrians in a long-term filmed surveillance environment,and has certain social application value.