Research On Intelligent Recognition Technology Of Vehicle Crossing Behavior Based On On-board Video

In recent years,with the acceleration of the urbanization process,China's transportation has been developing with each passing day,with the number of transportation means continuously increasing and the types of transportation being increasingly diversified,and the road infrastructure being constantly improved.According to the ministry of public security,by the end of 2019,China had 348 million motor vehicles,including 260 million cars,200 million private cars for the first time,and 66 cities had more than one million cars.Compared with 10 years ago,the number of motor vehicles has increased by more than 80%and the number of motor vehicle drivers has doubled.At the same time,the rapid growth of the number of motor vehicles has brought great convenience to people's travel.Meanwhile,the problem of road traffic safety has become increasingly prominent,and traffic governance has been paid more and more attention by the whole society.Among them,the construction of an intelligent traffic system combined with artificial intelligence technology(AI)has become an effective way to solve the problem of urban traffic governance.The automatic traffic violation identification system is an important part of intelligent traffic system,and its research focus is to realize the automatic identification and evidence collection of all kinds of traffic violations by comprehensive use of computer vision technology and artificial intelligence technology.This paper focuses on the automatic detection of vehicle line-crossing behavior,aiming at the problems of limited surveillance scope of bayonet camera and easy to be deliberately avoided by drivers,and proposes an intelligent recognition technology of vehicle line-crossing behavior based on the on-board video.The main research contents and research results include:(1)Aiming at the multi-vehicle detection efficiency in complex traffic scenarios,based on the Mask R-CNN neural network model,a real-time multi-vehicle detection algorithm in complex traffic scenarios was proposed by introducing CUDA(Compute Unified Device Architecture)parallel technology and optimizing the number of layers of model residual feature pyramid network in its ROI Align pooling operation.Through three groups of comparative experiments on the experimental data collected in the real traffic scene,the experimental results show that the algorithm can greatly improve the vehicle detection speed on the basis of ensuring the model detection accuracy.(2)Aiming at the problem of low detection rate of lane lines due to light change,vehicle shadow occlusion,lane line damage and other factors in complex road conditions,a real-time synchronous detection algorithm of multi-lane lines based on instance segmentation is proposed,and the effectiveness of the algorithm is verified by experiments.The algorithm firstly carries out semantic segmentation of traffic scene images to distinguish lane lines and background.On this basis,pixel points of discrete lane lines are clustered using the Mean Shift clustering method to form an example of each lane line.Finally,the pixel set of each lane line is fitted polynomially.(3)Based on the detection results of vehicles and lane lines,an automatic determination method is proposed to determine the violation behavior of vehicles.In this method,first of all,the vehicle coordinates obtained from vehicle detection are used as the benchmark to set the region of interest for a vehicle violation.Then,through the position relation between vehicle coordinates and lane line coordinates obtained from lane line detection,the illegal behavior of vehicles crossing the line is automatically judged.On this basis,a design scheme of intelligent recognition system for vehicle crossing behavior is proposed.