The Research On Lane Detection Algorithm Based On Image Processing

With the growing number of vehicles, the amount of traffic accidents increased year byyear and also resulted in a lot of casualties and significant property damage. In order to solvethe traffic problems, scientists have developed a variety of vehicle driving assist systems. Andthe basis component of all the systems is the lane detection. Incorrect lane identification willinterfere with the normal operation of systems. In recent years, domestic and foreign scholarshave made some significant lane detection algorithms, but two problems remained. The firstproblem is how to improve the algorithm’s robustness in a strong interference environment.The second is how to improve the real-time and do not affect the robustness in theenvironment.In order to solve the two problems, a fast and effective lane detection algorithm isproposed. Firstly, three image preprocessing methods, which include region of interestdivision, graying and image enhancement, are used to eliminate the noise introduced incapture stage and redundant information. According to the linear edge of lane, Gabor filter isapplied to highlight the lane. And in order to avoid the difficulty in determining a properthreshold in image segment on the complex road, Adaboost algorithm combined Haar featuresby some special rectangular features are designed for lane to extract the feature points of lane.In this process, an improved Adaboost algorithm is proposed to avoid defects of thetraditional Adaboost algorithm. Some fast processing methods are used in Haar and Adaboostprocesses to ensure the real-time performance without reducing the robustness. At last, animproved Hough algorithm is put forward to accelerate the calculation in extractingparameters of lane.This system is tested by a lot of experiments under various conditions. The accuracy ofthis feature extraction algorithm is about10%higher than the traditional algorithm, and theimproved Hough transform can save more than90%time-consuming. The results showed thatthis system can correctly identify the lane in real-time.