Vision-Based Drivable Surface Detection in Autonomous Ground Vehicles

One of the primary tasks for most autonomous ground vehicles is road following. For safe maneuvering the vehicle needs to correctly identify the drivable surface. Our work is focused on the use of simple video cameras as the sensor devices. We describe a new machine learning approach to drivable surface detection that automatically combines a set of rectangular features and histogram backprojection based image segmentation algorithms to produce superior results. The machine learning algorithm is based on the AdaBoost method, one of a class of boosting techniques which are applicable to many image processing tasks such as object and face recognition or image segmentation. The algorithm is trained and tested on video data obtained from video cameras mounted on an autonomous tractor at our Queensland site. The algorithm approach, together with the simple feature-based weak classifiers used, produces significantly improved drivable surface detection results