A Real-Time Detection Algorithm for Semi-structured Boundaries in the Park Based on 3D LiDAR

Efficient and accurate real-time detection of semi-structured driving boundaries within the park is required by unmanned ground vehicles (UGVs) in order to facilitate vehicle positioning and path planning. These boundaries have a regular shape, but do not have obvious line and surface features, such as low shrubs. It is crucial for ensuring optimal navigation for UGVs. We have developed a new method for detecting semi-structured boundaries based on 3-dimensional Light Detection and Ranging (3D LiDAR), which involves three key steps. The first step involved pre-processing the 3D raw laser point cloud, which included three tasks: coordinate correction, segmenting the region of interest (ROI), and ground point cloud segmentation to achieve point cloud down-sampling. Next, the non-ground points in ROI were projected onto the ground, and boundary feature points that affect the driving of UGV were extracted. To eliminate incorrect boundary points, a two-step method consisting of the Random Sample Consensus (RANSAC) algorithm and inter-frame boundary line angle elimination was used. In the third step, the data of Position and Orientation System (POS) were integrated to construct a global point cloud map of the semi-structured driving boundary. An experiment was conducted on a self-developed UGV-LiDAR platform to evaluate the performance of the proposed algorithm. It was indicated by the results of the study that the feature point detection rate was exceeded by 98.95%, and the average processing time per frame was around 45.7 ms when there are more shrubs, meeting the real-time demands. This research provided valuable insights into perception technology for security operations of UGVs in parks.