Vanishing Point Detection Based on The Fusion of Lidar and Image Data

Abstract

Despite the great progress that has been done in vanishing point-based methods for road detection using visual information the results are still vulnerable to external light conditions. For that reason, the fusion of LIDAR data, alongside images can be used for a more reliable result. LIDAR data may lack illumination information but are less susceptible to light conditions. The main contribution of this paper is the use of LIDAR data to create a mask that will restrict the area of the image that could eventually be the road. More specifically, we are using LIDAR data to detect the points of the ground. By contracting an Octree, we find the best-fitting plane of each leaf and by performing clustering we estimate the ground. Next, we are mapping the points of the road to the image to create a mask for the image processing step. We extract the texture orientation using Gabor Filter and thereinafter the vanishing point. The proposed approach has been implemented and tested with over 1000 images of different road scenes in the KITTI dataset. The experimental results demonstrate that this training-free approach can detect horizon and vanishing point very accurately and robustly, while achieving promising performance.