Research On Vehicle Violation Detection Method Based On Convolution Neural Network

In recent years,intelligent traffic system(ITS)has become the research hotspot of intelligent network industry,and it is indispensable to monitor the vehicle behavior on the road in its.Vehicle behavior monitoring is also vehicle detection,which mainly relies on a variety of sensors to sense the surrounding environment and vehicle information.At present,in vehicle violation detection,camera combined with loop coil detector and radar sensor is mainly used to detect vehicles,but there are defects such as false detection,missing detection and high cost.Therefore,in view of illegal vehicles on the road,this paper uses camera to detect illegal vehicles,The research was carried out for the detection technology.Firstly,this paper analyzes the current situation of vehicle detection technology of intelligent transportation system at home and abroad,proposes a multi-directional vehicle detection system based on convolution neural network,and integrates the multidirectional vehicle detection system and license plate recognition module to form a vehicle violation monitoring system.Secondly,the YOLO＿V2 network framework is improved,and the network structure is appropriately cut to improve the network training speed;the residual network structure is increased to improve the detection accuracy of convolutional neural network;the multiscale layer is added to improve the detection accuracy of vehicles of different sizes in the image;the activation function is designed to adapt to multi-directional vehicle detection.Thirdly,multi-directional vehicle data set and license plate data set are made.Due to the lack of research on multi-directional vehicle detection and the lack of available data sets,a multi-directional vehicle data set is made by using the Kitti data set and the road condition images collected by the dash cam,and a complete license plate data set and a single character data set are made respectively to complete the production of license plate data set.Finally,the multi-directional vehicle recognition network,license plate recognition network and vehicle tracking system are integrated in the ROS system(robot operating system),and the violation information of the detected vehicle is displayed through the QT creator visual interface.The violation monitoring system is transplanted into TX2 development board,and the hardware platform is used to verify under specific working conditions.The test results show that the vehicle detection based on convolutional neural network is feasible,which is of great significance to the development of vehicle behavior monitoring technology.