Research On Several Key Technologies Of Driving Assistance System Based On Stereo Vision

With the development of visual sensor and graphics and image processing algorithms, computer vision is becoming an indispensable part in the field of auxiliary driving. The vision-based drive assistance system uses camera as sensor, collects the image around the vehicle and analyzes it, it can provide the auxiliary and reminding for the driver, so as to ensure the driving safety of the vehicle. This article mainly focuses on the research of several key functions in driving assistance system, including obstacle detection and identification based on stereo vision, lane detection and modeling, lane tracking and identification of road surface,proposing improvements and attempts for some key technologies and difficulties in the driving assistance system.An obstacle detection method based on stereo vision is proposed for unstructured pavement. Dimensional information provided by the stereo vision to identify the obstacles in front of the vehicle. The system will promptly alert the driver for dangerous obstacles within the distance, and even make emergency braking and other driving behavior to assist vehicles safe driving.A road lane detection and modeling method is studied. This paper presents a hyperbolic lane model that can be used to describe different lanes with less parameters. In view of the lane line model is susceptible to light, block, shadow and so on, RANSAC algorithm is used to model the lane line, which enhances the robustness of the system and improves the modeling effect.The algorithm of lane tracking and pavement marking detection is completed. By using the kalman filter algorithm to track the location of the lane line and detect the lane change behavior of the driver in the time domain. It is compared with identification of road surface marks to remind the driver in accordance with normal traffic laws and regulations.Finally, the experiment on the whole system is completed. By using the open source data set to simulate the environmental information during the process of vehicle driving, the experimental results show that the proposed algorithm is effective and robust.