Design And Implementation Of Traffic Flow Information Acquisition Terminal Based On Convolution Neural Network Image Recognition

With the rapid development of the transportation industry,people are increasingly demanding urban traffic management.The traditional wave or magnetic traffic flow detection technology has become increasingly prominent in the areas of detection range,accuracy,and scalability.Computer video traffic flow detection technology can realize high-performance image processing based on the cloud,but there are disadvantages such as high cost,difficulty in installation and maintenance,and the like.To solve the above problems,this paper combines embedded IoT technology and artificial intelligence machine learning,and designs and implements traffic flow information collection terminal based on convolutional neural network image recognition.Based on the low-cost FPGA,YOLO fast target detection algorithm is adopted.Based on the features of traffic flow target detection and the high real-time requirements of the system,a streamlined re-inspection suppression mechanism and a degree of parallelism of 64 are designed.The convolutional hardware accelerator was designed and the network structure of YOLO was optimized and improved to realize traffic flow information collection for pedestrians,motor vehicles,and non-motor vehicles.Through the NB-IoT wireless communication module,the traffic information collected by the terminal is periodically transmitted to the cloud console,which assists the intelligent traffic system to complete intelligent management and control.This article is based on the low-cost ZYNQ-7000 series on-chip programmable system development platform,using PASCAL VOC 2012 as the network training data set,using software and hardware collaborative development methods to complete the system design and implementation.The road traffic video image with a resolution of 640*480 Piexl can reach 92.67% recall rate,98.72% precision rate,and 13.5f/s traffic speed target detection effect.It reaches the expected target.It provides an optional means for intelligent traffic detection with low cost,strong scalability and flexible deployment,and has broad market application prospects.