A Fast Spatial Clustering Method for Sparse LiDAR Point Clouds Using GPU Programming.

Area of Science:

• Robotics
• Computer Vision
• Geospatial Analysis

Background:

• Accurate obstacle detection is critical for mobile vehicle navigation and safety.
• Sparse and unstructured Light Detection and Ranging (LiDAR) point clouds pose challenges for traditional clustering algorithms, leading to inaccuracies and slow processing.
• Existing methods struggle with real-time processing of large 3D point cloud datasets.

Purpose of the Study:

• To develop a novel algorithm for fast and accurate obstacle clustering from LiDAR data in unknown environments.
• To leverage graphic processing unit (GPU) programming for accelerating the obstacle detection process.
• To enhance the environmental perception capabilities of mobile vehicles.

Main Methods:

• LiDAR points are projected onto a rasterized x-z plane, mapping sparse data into regularly arranged cells.
• Ground cells are filtered based on height distribution to generate a flag map.
• An Elevation-Reference Connected Component Labeling (ER-CCL) algorithm is applied to the flag map for cluster identification.
• Results are inverse transformed to 3D space, with parallel processing on a GPU for real-time performance.

Main Results:

• The proposed ER-CCL algorithm effectively clusters obstacles from sparse LiDAR point clouds.
• GPU acceleration significantly speeds up the 3D point cloud clustering process.
• The method demonstrates improved accuracy and efficiency compared to traditional approaches for obstacle detection.

Conclusions:

• The ER-CCL algorithm provides a robust and efficient solution for real-time 3D point cloud clustering and obstacle detection.
• This approach enhances the environmental perception capabilities of mobile vehicles, crucial for autonomous navigation.
• The integration of GPU programming is key to achieving the necessary speed for real-time applications.