3D-modeling techniques for object recognition based on point cloud data

This study examines 3D-modeling techniques for object recognition using point cloud data and evaluates the performance of key deep learning-based semantic segmentation techniques. This study compared AI structure for point clouds such as PointNet, PointNet++, PointCNN, and DGCNN (dynamic graph CNN). The comparison confirmed that each technique offers different approaches to effectively manage the characteristics and spatial structure of point cloud data. Notably, DGCNN demonstrated high segmentation performance through dynamic graph convolutional networks. This study provides essential foundational data that can contribute to advancing disaster response technology and is expected to significantly aid disaster response agencies in making quick and accurate decisions.