A Decision Tree Based Road Recognition Approach Using Roadside Fixed 3D LiDAR Sensors

Abstract

As one of the most important elements in the intelligent transportation system (ITS), the road traffic monitoring system (RTMS) needs to be functioned with a road recognition mechanism. Current works on road recognition mainly target at the field of automatic driving and cannot be directly used in the RTMS. In this paper, we propose a decision tree-based road recognition algorithm using roadside fixed light detection and ranging (LiDAR) sensors in the RTMS. These LiDAR sensors have a low vertical resolution, which implies that we cannot get a clear far boundary and obvious features of roads from the point cloud data. Point cloud data obtained by the roadside LiDAR sensors are projected onto a plane rasterized to grids of points. Using a decision tree, these grids are first classified into background grids and road grids. For reducing misclassification, these grids are further reclassified using a mean filtering algorithm. Finally, a minimum circumscribed rectangle algorithm is employed to obtain accurate road boundaries. The experiment results show that compared to existing road recognition algorithms, the proposed approach has advantages of being completely automatic, requiring shorter recognition time and having a wider detection range.