Research On Lane Detection And Departure Warning Methods For Assisted Driving

With the rapid growth of national car ownership,traffic accidents occur frequently,and traffic safety becomes increasingly serious.Active vehicle safety technology has become an important research topic.ADAS(Advanced Driving Assistant System)is the key to achieve active vehicle safety,and lane departure warning is an important function of ADAS.In this paper,the hardware structure and software algorithm of the lane departure warning system are systematically analyzed based on machine vision and mechanical design.The camera calibration method,lane detection and tracking method and departure warning strategy are studied deeply.This study provides theoretical and technical support for active safety and driverless vehicles.The research in this paper firstly starts from the hardware system.The system adopts the vehicle-mounted monocular camera to collect the road image in front of the vehicle in real time.The camera is selected according to the system requirements.The main structures of the phase camera module are designed on the process requirements of ensuring the imaging quality.Then in the part of software algorithm,the calibration method of internal and external parameters of vehicle camera is studied.The static plane target was used to calibrate camera parameters.According to the principle of perspective projection and the model of small hole imaging,a method of vehicle camera external parameter calibration using vanishing point and geometric characteristics of lane is proposed.The method can estimate and modify the external parameters of the camera in real time in the process of vehicle driving,so that it can adapt to the change of external parameters caused by road turbulence or vehicle load,and avoid the complex process of traditional calibration method.Lane detection and tracking is the key technology of lane departure warning.A method of lane line detection based on edge feature point clustering is proposed for complex road conditions and real-time requirements of the system.Edge feature points are extracted according to the distribution features of lane edge gradient and gray scale.Then clustering is conducted according to the continuity of lane and the consistency of feature point gradient direction,and regional clustering is conducted by measuring the similarity of each region.The optimal class is selected for RANSAC linear fitting,and according to the fitting results and Kalman filter dynamic prediction detection area,the lane is quickly detected and accurately tracked.This paper analyzes the characteristics of the common decision models of lane departure,and puts forward the ETLC decision model based on the driver’s driving behavior and vehicle movement trend.Finally,the above methods tested on the simulation platform and the actual road.The experimental results show that the system meets the requirements of ISO 17361 and is of great significance for the engineering and productization of the auxiliary driving system.