Pedestrian Attribute Recognition System Development Based On Deep Learning

Pedestrian attribute recognition has obvious characteristics such as intelligence,precision,and real-time.Its research and application have important practical significance for pedestrian identification,pedestrian retrieval,and personalized service applications in the fields of commerce,security,and urban management.This paper takes pedestrian attribute recognition system as the research object,combined with the increasing popularity of video surveillance and the urgent need for intelligentization,this paper designs and implements a deep learning pedestrian attribute recognition system for monitoring scenarios with potential application value of pedestrian attribute recognition such as banks and shopping malls..Based on the analysis of the needs and feasibility of the system,the overall architecture and composition of the system,work flow and interactive interface are given.The system involves pedestrian detection and pedestrian attribute recognition,related object detection network and attribute feature extraction Network and other aspects.In the system design,RFBNet was compressed and improved on the network structure to improve the performance of pedestrian detection;based on the Market-1501 data set,a multi-attribute classifier network based on Res Net50 to extract features was designed to achieve the pedestrian attribute in the system.Recognition;The adjacent frame enhanced attribute recognition technology based on the attribute recognition model is proposed to improve the recognition accuracy of the system.Based on the design of the pedestrian detection model and pedestrian attribute recognition model,the system was implemented,and the system was evaluated for accuracy and speed.Among them,the improved RFBNet parameter model compression rate reached 10%,and the pedestrian detection model The accuracy of the and attribute recognition models reached 0.73 and 0.90 on their respective test sets,and the system recognition speed reached 1.6fps.In general,a pedestrian attribute recognition system with remarkable characteristics such as high precision and real-time is designed and implemented,which lays the foundation for the development of more pedestrian attribute recognition systems in the future.