Point Cloud Segmentation Algorithm on Convolutional Neural Network

With the rapid development of 3d point cloud acquisition methods, point cloud segmentation has become increasingly important in computer vision, logistics robot and other fields. We proposes a point cloud segmentation method on convolutional neural network to realize semantic segmentation of indoor point cloud scenes, which has uneven density, complex object category, chaotic spatial direction and no pixel information. This method can extract local information, which mainly including nearest neighbor normal vector, curvature and Euclidean distance, global information supplemented by spatial height from all points in point cloud. The convolutional neural network is used to train the extracted information and to realize the classification of objects. The experimental performance and effectiveness has been verified by simulation and experiments, obtain 0.17 mAP@0.5 on ScannetV2.