EXPLORING GROUND SEGMENTATION FROM LIDAR SCANNING-DERIVED IMAGES USING CONVOLUTIONAL NEURAL NETWORKS

Abstract. Recent works have attempted to extract features such as road markings from sparse mobile LiDAR scanning point cloud-derived images via convolutional neural networks (CNN). In this paper, the use of methods for ground segmentation was explored. To begin, clouds each channel will be projected onto y-z plane generate that used training and testing CNN model. Then, main workflow, following steps were performed channel: (1) cloud-to-image conversion; (2) classification; (3) image-to-point cloud projection. Then utilizing multi-threading, is processed in parallel our ground-segmented cloud. Our findings shown successful segmentation, achieving an f1-score 98.9%. However, it 27.81% slower compared RANSAC. Overall, initial investigation has demonstrated imagery possible, with further improvements model, make faster, good potential act alternative conventional processing.