Application of Multi-Feature Fusion Based on Deep Learning in Pedestrian Re-Recognition Method

Abstract

A system known as pedestrian recognition makes use of several cameras to identify the surrounding area and quickly identify and match the target demographic. Based on pedestrian recognition, the picture model, pedestrian features, and other information, the features are developed to have a high degree of generalizability, distinctiveness, and accuracy. The application approach for pedestrian re-recognition based on deep learning for numerous features is proposed in this paper. The suggested approach successfully preserves high-level semantic information, which helps network members extract all of the pedestrian properties. As external material and semantic information were combined horizontally and vertically, environmental interference was decreased, and people’s ability to create networks was enhanced. The voice channel of the speech system was introduced in order to fully utilize the global information network, and the connection between the channels was carefully addressed in order to enhance the global information network’s capacity for expression. The null convolution reduced the operational continuity of the identification information. To increase the consistency of the data, the multi-level spatial convolution structure was merged with the entire image in this paper. After numerous experiments, the three groups were 89.5%, 89.5%, and 89.1%, respectively, compared to 1501, DukeMTMC-reID, CUHK03, and other medial groups, and the experimental results were 85% and 89.5%, respectively. The multimode feedback MP3 module was taken from the MP3 module in order to gain richer and denser multimode feature information. Comparing the module’s initial response level (RANK1) with the various cycles yields the average accuracy for each cycle (catalog). The experiment demonstrates that the two mixed pile groups can enhance the modulus of the mixed pile group and get better results. The multi-level multi-scale pole function effectively combines the characteristics of pedestrians in various scales, and the addition of the ASP module enhanced the network context information’s overall ability to be represented, aided in this chapter’s research method’s ability to more thoroughly analyze scene structure, and increased the precision of pedestrian re-recognition.