2019/07/03 by Florian Piewak, Piewak, Florian, Peter Pinggera +3
Computer Science · Engineering · #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Infrastructure Maintenance and Monitoring #Machine Learning (cs.LG) #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.1907.02149
openalex publication_date 2019/07/03 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
State-of-the-art approaches for the semantic labeling of LiDAR point clouds\nheavily rely on the use of deep Convolutional Neural Networks (CNNs). However,\ntransferring network architectures across different LiDAR sensor types\nrepresents a significant challenge, especially due to sensor specific design\nchoices with regard to network architecture as well as data representation. In\nthis paper we propose a new CNN architecture for the point-wise semantic\nlabeling of LiDAR data which achieves state-of-the-art results while increasing\nportability across sensor types. This represents a significant advantage given\nthe fast-paced development of LiDAR hardware technology. We perform a thorough\nquantitative cross-sensor analysis of semantic labeling performance in\ncomparison to a state-of-the-art reference method. Our evaluation shows that\nthe proposed architecture is indeed highly portable, yielding an improvement of\n10 percentage points in the Intersection-over-Union (IoU) score when compared\nto the reference approach. Further, the results indicate that the proposed\nnetwork architecture can provide an efficient way for the automated generation\nof large-scale training data for novel LiDAR sensor types without the need for\nextensive manual annotation or multi-modal label transfer.\n