Design And Implementation Of Dynamic Traffic Violation Detection System Based On Machine Vision

With the continuous development of industry and the continuous improvement of economic level,the number of cars has increased significantly,which increases the frequency of traffic accidents and traffic violations.In order to curb the deterioration of this situation,China introduced and developed the traditional static traffic violation detection system "electronic police" earlier,but it has many shortcomings,such as fixed camera,limited detection range,limited illegal acts captured and so on.In this context,thesis designs a vehicle embedded traffic violation detection system based on machine vision.The research work of thesis includes the following aspects.Design the overall architecture of the system.At the hardware level,thesis selects the ARM embedded processor hi3559av100 as the basis of the system hardware platform;At the software level,thesis designs a star network topology centered on the message processing module,encapsulates different functional blocks into modules to form each node of the star network structure,and takes it as the network architecture of the software system.Based on the system network architecture,six functional modules are designed and implemented: message processing module,media information processing module,storage module,system module,human-computer interaction module and algorithm module.Based on the star network architecture,thesis realizes the communication between modules by using TCP message protocol,and establishes the media information data channel based on shared memory allocation technology.In addition,MCU,Wi Fi module,Linux instruction script,search tree and other related technologies are used to build the system,support relevant functions,and ensure the stable operation of the system.In terms of violation detection algorithm,the system realizes vehicle target detection based on RetinaNet algorithm,vehicle target tracking based on DeepSort algorithm,and detects and classifies lane lines using a priori deep learning lane line detection method.On this basis,according to the detected relative position information between the target vehicle and the lane line,a specific road violation detection algorithm is designed to judge whether the target vehicle has violations such as pressing the Yellow / white solid line,changing lanes without turning lights and retrograde,and capture and save the traffic violations.Finally,the function verification and performance test of the system are carried out.The verification shows that the dynamic traffic violation detection system software designed in thesis can realize the functions of video playback,data storage and humancomputer interaction,and operate stably.The algorithm can detect,capture and save the expected road traffic violations in real time,which meets the design requirements of thesis.