Development of Vehicle Detection Method on Water Surface Using LiDAR Data for Situation Awareness

Abstract

In order to operate unmanned autonomous surface vehicle, it is necessary to strongly detect water obstacles. Among the sensors for this purpose, LiDAR sensor is the most intuitive at close range and can detect obstacles strongly regardless of surrounding environmental conditions. In order to detect obstacles with 3D LiDAR sensor data, clustering of point clouds for each object is first required. These clustered point groups are used to classify the types of objects. In this study, a convolutional neural network is used to classify the types of objects. Since 3D point cloud data cannot be directly entered into this network, we propose the descriptor that can express the representative characteristics of the clustered point cloud. Using this descriptor, 3D point cloud data can be converted into a 2D image, and the converted 2D image is provided as an input value of the network. Using the experimental results on the simulator, we intend to verify the validity of the use of the point cloud feature descriptor proposed in this study.