Research On Preceding Vehicle Detection System Of Intelligent Vehicle Based On Convolutional Neural Network

At present,intelligent vehicle has become a research hotspot in the automobile industry,and it is a new industry combining industrialization with information technology.Intelligent vehicle mainly relies on a variety of sensors for environmental information sensing to make further decisions,among which vehicle detection technology is one of the most important perception technologies in environmental information perception system.This paper mainly studies the preceding vehicle detection technology of intelligent vehicle.Firstly,this paper investigates and analyses the development status of vehicle detection technology at home and abroad,and proposes a vehicle detection system based on deep convolution neural network,and divides it into three modules: image acquisition module,image processing module and information output module based on modular design concept.Secondly,based on the YOLO network framework,the vehicle detection convolution neural network model in this paper is designed by tailoring its structure properly and modifying its scale adaptability to improve the real-time performance of the network and the detection ability of small targets.And the two vehicle detection data sets,KITTI and Udacity,are merged and modified,and by adding domestic vehicle data sets,the final training sample set is made.Dimension clustering of label boxes in the vehicle data set is carried out by K-means++ algorithm to obtain representative anchor parameters to accelerate the convergence speed and increase the convergence degree of the network.When the network model is built and parameters are set up,the network is trained in stages on the vehicle training sample set until the network converges.Thirdly,in order to further improve the accuracy and reliability of the whole vehicle detection system,a vehicle tracking algorithm based on Kalman Filter is designed based on the detection results of vehicle detection network,and the detection results and tracking results are fused.The vehicle trajectory is predicted by using vehicle position coordinate sequence saved in vehicle tracking algorithm,and the relative deceleration of the front vehicle and the lateral vehicle insertion situation are selected for judgment and detection.Finally,the vehicle experimental platform is built including the selection and parameters calibration of the camera and the vehicle ranging by monocular vision.The designed program on PC is transplanted to the embedded development platform Jetson TX2,and the vehicle detection system is validated by the real vehicle experimental,and the experimental results verify the effectiveness of the proposed preceding vehicle detection system of intelligent vehicle based on convolutional neural network.