Fig. 1. Projection type and the depth map

Fig. 2. Distance feature projection image

Fig. 3. Schematics of angle feature. (a) Vertical angle feature α; (b) horizontal angle feature β

Fig. 4. Angle feature images and their effect after filtering. (a1) Vertical angle feature image; (a2) vertical angle feature image after filtering; (a3) comparison of pixel values before and after filtering of a column of pixels; (b1) horizontal angle feature image; (b2) horizontal angle feature image after filtering; (b3) comparison of pixel values before and after filtering of a row of pixels

Fig. 5. Projection image of curvature feature

Fig. 6. Three-dimensional matrix containing primitive information and feature information of point cloud

Fig. 7. Structure of multi feature point cloud convolution neural network

Fig. 8. Comparison of semantic segmentation effect among the proposed method with SqueezeSeg V2 and PointSeg, as well as the ground truth and images of the corresponding scene images

Fig. 9. Comparison of running performance