Research On Detection Algorithm Of Road Traffic Markings

With China’s rapid economic development, more and more families have cars, and so the problem of traffic safety has become increasingly serious. In order to improve the safety of drivers and passengers, safety driver assistance systems and intelligent vehicles have become a hot research topic in recent years. Intelligent vehicles in the case of unmanned need the perception of the external environment, and road markings are important information in the external environment, so road marking detection has very important significance for intelligent vehicles. To ensure the safe driving of intelligent vehicles, the technology of road marking detection detects road markings in real-time by obtaining the road scene in front of intelligent vehicles.In this thesis, road markings are detected by road images, which are obtained by the monocular camera installed on the intelligent vehicle. Detection objects include lanes, crosswalks and stop lines. The main research contents are as follows:(1) The inverse perspective mapping method of the image is studied. The transformation between the vehicle body coordinate system and the image coordinate system is analyzed. The conversion formula between the image pixel coordinate and the actual physical distance is calculated. And the inverse perspective mapping is carried out on the region of interest in the road image.(2) The threshold segmentation algorithm based HSI is improved. Combining with OTSU, road markings are segmented in H、S and I space respectively. The binary image of the road is obtained by the integration of the image segmentation and edge detection information.(3) The lane, crosswalk and stop line detection algorithms are studied. The rectilinear coordinate on the binary image is obtained by using the improved probability Hough transform, and then lanes are detected after feature matching. A crosswalk detection algorithm is designed based on bipolar and feature fusion. The crosswalk candidate region is extracted by the bipolar, and the multi-feature fusion is carried out based on the characteristics of the crosswalk width, height and angle, so that crosswalks are detected. A stop line detection algorithm based on grayscale image is presented. The feature points of the stop line are searched by the convolution filtering with the selected template nuclear, and then they are matched by using the width invariance of the stop line, so as to get the candidate region of the stop line, and finally the stop line is detected by the Hough transform on this candidate region.The experimental results show that, the presented road marking detection algorithm possesses strong applicability, and can be applied to a variety of environmental conditions, and be satisfactory on the real time request.