Sloped terrain segmentation for autonomous drive using sparse 3D point cloud.

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Geospatial Data Processing

Background:

• Autonomous driving requires precise real-time localization, which is challenged by the need to minimize data exchange in wireless environments.
• Accurate ground segmentation is crucial for obtaining reliable location data from vehicles operating in complex terrains.

Purpose of the Study:

• To propose a novel framework for real-time ground segmentation using sparse 3D point clouds.
• To enable efficient data processing for autonomous vehicles navigating undulating terrain.

Main Methods:

• Utilizing light detection and ranging (LiDAR) sensors to acquire sparse 3D point clouds.
• Quantizing point clouds into voxels and eliminating overlapping data for efficient processing.
• Implementing a lowermost heightmap to reduce voxel data to two dimensions.
• Determining ground area based on voxel group analysis and minimizing neighbor comparisons for real-time execution.

Main Results:

• The proposed framework achieves real-time ground segmentation.
• Experimental verification shows the algorithm operates at approximately 19.31 milliseconds per frame.
• The method effectively segments ground in sparse 3D point clouds from undulating terrain.

Conclusions:

• The developed framework provides an efficient solution for real-time ground segmentation in autonomous driving applications.
• This approach facilitates accurate localization and reduced data exchange, crucial for ubiquitous road travel environments.
• The algorithm's speed and accuracy make it suitable for simultaneous autonomous driving operations.