Fast reducing perspective distortion for road marking recognition

Min Feng, Bin Fang, Weibin Yang

Abstract: Road images taken from autonomous vehicles usually have perspective effects, which distort the shapes of road markings and hence make the subsequent road marking recognition difficult. A commonly used method to solve this problem is the Inverse Perspective Mapping (IPM), which could alleviate or even eliminate the perspective effect. However, it is time-consuming due to the interpolation operation. In this paper, we propose an efficient algorithm to reduce the perspective distortion of road markings. First, we reconstruct the road images by changing the viewpoint, as if they are captured by the camera right in front of the road markings in each lane. Second, Region of Interest (ROI) of each lane could be confirmed in its corresponding image. The proposed method could provide a fast and stable view of road markings and make them easy to detect and recognize. The experiments on a road image dataset demonstrate that our proposed algorithm is much faster than IPM while maintaining the comparable accuracy in road marking recognition.