Research On Lane And Vehicle Detection Based On Vehicle Vision

With the continuous advancement of deepening reforms,China's gross domestic product has repeatedly hit new peak.While the standard of living is constantly improving,the number of car ownership in China has increased year by year,and various types of traffic accidents have occurred frequently,seriously affecting the personal and property safety of the people.In order to reduce the damage caused by traffic accidents,scientific research institutions and enterprises are increasing their personnel and capital investment to study the intelligent driving system of intelligent vehicles,including vehicle departure warning and vehicle collision warning.Vehicle-assisted driving and intelligent driving systems rely on the detection of car environment information.Among many sensors,vision sensors are favored by researchers at home and abroad due to their low cost and abundant color information.This thesis uses BJUT-IV intelligent vehicle of Beijing University of Technology as a hardware platform to study the lane line and vehicle detection technology based on vehicle vision,including coordinate transformation between image coordinate system and world coordinate system,image distortion correction,fast lane line identification in curved road based on optical flow,vehicle identification based on neural network and distance detection based on binocular matching,and lane departure and collision warning system design.The main work of this thesis is as follows:1)According to the imaging model of the vision sensor,the coordinate transformation method between the image coordinate system and the world coordinate system is studied,and the image distortion is corrected by the camera calibration method.2)For the problem of lane line recognition under curved road conditions,a corner detection method using multiple gray-scale operators is proposed.The optical flow method is used to dynamically determine the region of interest of lane line detection and speed up lane recognition.First,based on the temporal correlation between successive video frames,an optical flow algorithm is used to detect the relative movement of the background in front of the vehicle.Then,using the moving direction and distance of the feature points in the background of the front of the vehicle,the noise is screened out to obtain the overall offset vector,and the position of the lane line in the frame image is roughly positioned to narrow the detection range.Finally,a multi-scaled operator is used to obtain the binarized image,and the least squares method is used to fit the lane line of the curved road.The method can detect the position of the lane line under the structured road in real time and accurately.3)To solve the problem of vehicle identification in front of intelligent vehicles,a forward vehicle identification method based on convolutional neural network is presented.Aiming at the difficulty of modeling deep neural network structure,this thesis studies the learning method of automatic design neural network.By designing the rules and parameter migration methods of network automatic learning,combined with stochastic gradient descent algorithm to construct neural network,the recognition rate of the generated convolutional is superior to the traditional detection algorithm and the Faster-RCNN network.4)The lane departure warning is detected by detecting the distance between the lane line and the vehicle;using the binocular matching method,the distance of the preceding vehicle is detected by parallax calculation and three-dimensional reconstruction,and the vehicle collision warning is performed.Finally,according to the needs of the subject,a vehicle vision assisted driving system was compiled to verify the practicability of the algorithm.