Research On Passenger Flow Detection And Tracking System Based On Edge Computing

In recent years,with the rapid development of artificial intelligence and the maturity of target detection technology,various systems based on pedestrian detection technology have been widely used in areas such as unmanned driving,intelligent transportation,and intelligent monitoring.Using computers instead of human eyes can improve the efficiency of target detection on the basis of greatly reducing the consumption of manpower and materials,which makes the passenger flow detection system in the focus and focus of research in the field of security monitoring.However,the existing passenger flow detection systems often rely on the server for development and application,which makes the background operation pressure increase,the calculation amount increases,and the bandwidth cost increases.Based on the above issues,this article quotes the edge computing idea,transplants target detection,tracking,passenger flow data statistics and other tasks to front-end edge devices for development and application.It studies a small detection network suitable for edge computing and selects an embedded GPU platform to complete the overall system.Build.The main research results are as follows:For the pedestrian detection module,this paper proposes a lightweight pedestrian detection algorithm based on the YOLOv3-tiny network,and uses the deep separable convolution to reconstruct the Yolov3-tiny backbone network.On the premise of ensuring the algorithm detection speed,the network's ability to extract pedestrian features is improved.,And add a layer of prediction to ensure that targets of various sizes are accurately detected,and to ensure the accuracy of network detection.For the pedestrian tracking module,this paper proposes a target tracking algorithm based on deep learning and multi-feature fusion.Using the pedestrian target position information obtained by the pedestrian detection module,the multi-feature extraction of the pedestrian target that appears in each frame of the video to be detected can be extracted and based on the extraction The obtained multi-features use Hungarian algorithm for data association,and finally form a tracking trajectory.This paper selects the NVIDIA TX2 development board as the system development platform,and researches and develops modules for pedestrian detection,target tracking,and data analysis that are suitable for the platform.At the level of software architecture,the functions of passenger flow statistics such as pedestrian target detection,multi-target tracking,pedestrian number,density,speed,and direction of travel are realized.The passenger flow detection system has been tested for system stability,functionality,and accuracy in multiple subway scenarios.The experimental results show that the system's stability and functionality have met the project requirements,and the system detection accuracy rate has reached 95.8%.,The operating frame rate reaches 18 fps,which basically meets the application needs of real-time subway scenes.