Research And Testing On Lane Detection And Tracking Algorithm Based On Monocular Vision

Nowadays transportation is facing huge challenges of accidents and congestion with the rapid growth of vehicle numbers. The research and development of efficient, safety and comfortable driverless intelligent vehicle has been given considerable attention by scholars and enterprise from domestic and abroad. Where the environment perception based on visual is the key technology of intelligent vehicle to realize unmanned. Lane detection and tracking is the essential function for intelligent vehicle to correctly identify the road region can travel. Therefore, the ability to use real-time visual sensors to sense road conditions and accurately detect and track the lane is a prerequisite for calculating the travel region of the unmanned intelligent vehicle.This thesis is based on vision and focusing on how to achieve accurate real-time detection and tracking algorithm under complex structure road lane, carried out research and testing. The main contents are as follows:Firstly, preprocessing was performed to obtain an image of real road locations, simplified acquired image data to enhance the lane detection and tracking algorithm accuracy and timeliness. Secondly we detected and tracked lane of road image and used detection algorithm to extract lane marking positive and negative edge points. Based on Hough transform voting mechanism to extract edge of a line, and through line constrain, matching the edge lane and marking line to realize lane detection. Tracking algorithm is based on the filtering forecast and update state parameter of the lane line model, so as to track the lane line trajectory and fit the lane line. Finally, adaptive improve lane detection according to self-limited of visual sensor, past lane trajectories and high-precision map.A large amount of field measurements of has been taken through intelligent vehicle to measure the performance of the algorithm, measurement contains of a variety of complex condition of structured road scenes. The results show that lane detection and tracking algorithm can deal with different challenges, including complex patterns, vehicle occlusion and ground noise. The real-time and robustness of lane detection algorithm meet each indexes of the evaluation system of the algorithm, achieved the requirements of the lane recognition for intelligent vehicles.