Lidar Image Classification based on Convolutional Neural Networks

This paper presents a new method of recognition of lidar cloud point images based on convolutional neural network. This experiment uses 3D CAD ModelNet, and generates 3D point cloud data by simulating the scanning process of lidar. The data is divided into cells, and the distance is represented by gray values. Finally, the data is stored as grayscale images. Changing the number of cells dividing point cloud results in different experimental results. Experiments show that the proposed method has higher accuracy when dividing the cloud with 27 × 35 cells. Comparison of point cloud cell image method with VoxNet method, experimental results show that the classification method based on gray image and convolutional neural network has more advantages than the most advanced point cloud recognition network Voxnet.