Multi-feature Extraction for Drivable Road Region Detection with a Two-Dimensional Laser Range Finder

摘要

A multi-feature extraction method using a single 2D Laser Range Finder is proposed to provide the information of drivable road region for the Unmanned Ground Vehicle in structured environment. The whole process consists of two main steps: first, the curbs are extracted from the range data using the wavelet transform method, and evaluated by a road scanning model; then, the drivable region is detected based on the average height and flatness. A mathematical model for curb and road detection based on grid map is built to guide the vehicle. The algorithm distinguishes itself from the most existing methods which detect curbs only, it takes advantage of the flatness of surface between extracted curbs is employed as another feature for drivable road region detection. Furthermore, laser intensity data from the same laser range finder is introduced to supply a new restriction for surface evaluation. Experimental results verify the feasibility and stability of the method.