Target Detection from 3D Point-Cloud using Gaussian Function and CNN

This paper proposes a roadside-LiDAR-based target detection method using Gaussian probability function and CNN. First of all, the point-cloud is projected by the orthographic projection method to obtain the feature information on three projection plain. Then the Gaussian probability function is used to convert them into three probability matrices as the three-channel input of convolutional neural network. Finally the targets are divided into three categories: pedestrian, motor vehicle and non-motor vehicle. The experiments are performed with real data collected from different times and different scenarios, the results show that the proposed method can detect targets accurately, and the accuracy can reach 90%. The method is independent of the point cloud density, it has the same effect on sparse or dense point clouds.