A GPU-accelerated LiDAR Sensor for Generating Labelled Datasets

The reconstruction of tree branching structures is a longstanding problem in Computer Graphics which has been studied over several data sources, from photogrammetry point clouds to Terrestrial and Aerial Laser Imaging Detection and Ranging technology. However, most data sources present acquisition errors that make the reconstruction more challenging. Among them, the main challenge is the partial or complete occlusion of branch segments, thus leading to disconnected components whether the reconstruction is resolved using graph-based approaches. In this work, we propose a hybrid method based on radius-based search and Minimum Spanning Tree for the tree branching reconstruction by handling occlusion and disconnected branches. Furthermore, we simplify previous work evaluating the similarity between ground-truth and reconstructed skeletons. Using this approach, our method is proved to be more effective than the baseline methods, regarding reconstruction results and response time. Our method yields better results on the complete explored radii interval, though the improvement is especially significant on the Ground Sampling Distance In terms of latency, an outstanding performance is achieved in comparison with the baseline method.

show abstract

“…According to the LiDAR simulation, it is performed using the work described by López et al (2021a), López et al (2022) since it…”

Section: Synthetic Lidar Datamentioning

confidence: 99%