Research On Driving Environment Perception Technology Based On Vehicle Vision

Recently,better traffic capacity and traffic safety can be implemented using computer control,artificial intelligence and communication technologies.The lane departure warning system,avoidance collision warning system,intelligent cruise control and other functions in the intelligent driving system can greatly improve the intelligence of the vehicle and improve the safety of the transportation system.Visual perception is one of the most critical technologies,as all important decisions made by autonomous vehicles rely on visual perception of the surrounding environment.Based upon the perception outcome,an intelligent system can further make decisions to control and manipulate the vehicle.In this paper,machine learning and computer vision algorithms are often used to process the sensing data and derive an accurate understanding of the surrounding environment,including vehicle and pedestrian detection,traffic sign detection,etc.The main work of the paper is as follows:(1)Aiming at the problem of large size difference of targets in driving environment,this paper introduces SPP(Spatial Pyramid Pooling)module in YOLOv3 network,and designs the detection algorithm of road targets(vehicles,pedestrians,traffic signs)based on YOLOv3-SPP.The network resolution is adjusted according to the image size of the training set,and the value of the initial anchor box is reclustered.At the same time,the number of network detection categories is adjusted to transform the multi-category detection and classification problem into a three-category detection and classification problem for vehicles,pedestrians and traffic signs in the driving road scene.(2)In order to solve the problem of low recognition accuracy caused by factors such as illumination,object occlusion and other factors,A traffic sign recognition algorithm based on caps Net is designed.This paper expounds the defects of traditional convolutional neural network and the basic structure of CapsNet network,and based on CapsNet with inception branch to adjust the network structure.Data-enhanced traffic signs are put into the network model for training,and the network parameters are adjusted accordingly to make the accuracy of the model in the test set optimal.Finally,the python interface is used to call the above design modules under different frameworks to complete the design of environment perception algorithm under vehicle vision and realize the function of simultaneously detecting and classifying vehicles,pedestrians and 43 types of traffic signs.The target detection algorithm designed in this paper can accurately locate the road target and achieve a good balance between accuracy and speed.On the German traffic sign data set,the recognition effect of traffic signs based on CapsNet is compared with the recognition effect of traditional convolutional neural network.Experimental results show that the recognition rate of traffic sign classification model based on improved CapsNet is 98.73% and the convergence rate is faster.It has higher robustness and accuracy in complex environments.