Self-supervised Monocular Road Detection in Desert Terrain

Abstract

Abstract-We present a method for identifying drivable surfaces in difficult unpaved and offroad terrain conditions as encountered in the DARPA Grand Challenge robot race. Instead of relying on a static, pre-computed road appearance model, this method adjusts its model to changing environments. It achieves robustness by combining sensor information from a laser range finder, a pose estimation system and a color camera. Using the first two modalities, the system first identifies a nearby patch of drivable surface. Computer Vision then takes this patch and uses it to construct appearance models to find drivable surface outward into the far range. This information is put into a drivability map for the vehicle path planner. In addition to evaluating the method's performance using a scoring framework run on real-world data, the system was entered, and won, the 2005 DARPA Grand Challenge. Post-race log-file analysis proved that without the Computer Vision algorithm, the vehicle would not have driven fast enough to win.

Self‐supervised learning to visually detect terrain surfaces for autonomous robots operating in forested terrain

“…Dahlkamp et al (2006) apply a Kalman filter to estimate the robot pose based on measurements from two differential GPS systems and a six-degree-of-freedom inertial measurement unit (IMU). Based on the pose estimation, 2D LIDAR points are projected into a 3D frame.…”

Finding multiple lanes in urban road networks with vision and lidar

“…Since our system uses road paint as its primary information source, in the absence of road paint it is no surprise that no lane estimate ensues. Other environmental cues such as color and texture may be useful in such situations (Dahlkamp et al 2006). …”

Fast color/texture segmentation for outdoor robots

“…Other work has previously been done in road detection where we mention the work by Fernandez and Price [18] who used region growing in HSI color values to find the road borders. Others include Dahlkamp et al [19], who used self-supervised learning on color images to extend roads found by LADAR. In the area of stereo-vision Soquet et al [20] used a stereo-based color segmentation algorithm to determine road segments.…”

“…These can be computed recursively (18)(19) and are constrained to be nonnegative. The new width profile, w R i , of row i is computed as and takes into account spatial coherency in each row.…”