Research And Implementation Of Lane Detection System

With the explosive development of the domestic transport, many people were killed in traffic accidents because of negligent driving in every year. Intelligent transportation driver assistance systems are focused on reducing such traffic accidents. Lane detection system as a key part of the intelligent transportation driver assistance systems has been widely used in vehicles autopilot system, automatic vehicle parking system, vehicle anti-collision warning systems. Lane detection system uses digital image processing and pattern recognition techniques in road images to efficiently extract lane information and fit lane. This paper focuses on the realization of lane detection system and the theoretical knowledge focuses on pre-processing algorithms, conventional lane detection, lane detection algorithm based on the template of the study.(1) Preprocessing algorithms. This innovation puts forward the preprocessing algorithm RMC(Use Retinex and Mask to improve test results got by Canny). Firstly, we use the multiscale Retinex algorithm to enhance the original image, and use the bottom of the enhancement image to calculate a threshold value. Then we can obtain a binary image I1. Secondly, we can use the binary method based a mask to obtain a binary image I2. Thirdly, we use canny edge detection operator to obtain an edge image I3. Finally, we can get the preprocessing image through I1, I2 and I3. The algorithm can effectively remove the shadow image, and reserve the complete information of the lane in I1.It can not only effectively preserve lane information, and can remove the area where color changes slowly in I2. And I3 can get the original image edge information effectively. RMC algorithm can effectively remove large amounts of invalid information, and reserve the basic integrity of the lane edge information, which will reduce amount of calculation in the next section.(2) Conventional lane detection algorithm. In this paper, we put forward an improved algorithm CHEVP which is based on the Canny / Hough Estimation of Vanishing Points(CHEVP), and use this model to complete lane detection. Firstly, the image is divided into N regions, and use the Hough transform to obtain all line segments in each region. Using voting principle calculates the vanishing line, and through the vanishing line we can get vanishing points and lane line segments within each region. Finally, using the control points fits the lane. In this paper, Gauss-Like Model can improve Hough Transform defects, and calculated the coordinates of the vanishing line by weighted linear segments. This method is more robust and a higher accuracy rate than the previous calculation method.(3) Lane detection algorithm based on a template. Based on the tracking algorithm CSK(Exploiting the Circulant Structure of Tracking-by-Detection with Kernels) we put forward an improved tracking algorithm CSK. This algorithm is an effective solution to the shortcomings of original CSK which is the feature single and cannot handle the shade. This part can improve timeliness and robustness of the entire system.All the algorithms presented in this paper are tested and validated on VS2010 platform. Through using random sample test experiments, we can prove that the algorithm has strong robust, high accuracy rate and other characteristics.