Journal of Engineering Research

In order to more effectively obtain road surface information in the traversable area and improve the performance of three-dimensional laser radar point cloud feature extraction and point cloud segmentation, this paper proposes a road surface detection algorithm based on polar coordinates. Compared with Cartesian coordinate system segmentation, the use of polar coordinates to process the three-dimensional data generated by the laser radar can achieve a more uniform and accurate segmentation effect. At the same time, the introduction of two-dimensional CNN when segmenting the point cloud can greatly reduce the time consumption. Then the obtained point cloud is input into PointNet for grid point cloud feature extraction and RingCNN is introduced, and finally a good road surface detection accuracy is obtained, providing safe and reliable drivable area information for autonomous driving.

Traversable area; Road surface detection; Deep learning; Feature extraction; Point cloud segmentation