LiDAR Curb Detection Using Multi-Plane Slope Verification

The patent segments the curb detection process into multiple distinct stages: initial curb candidate point detection using heuristic algorithms, followed by verification steps that check geometric constraints (constant slope in plan view, constant sign slope in profile view), and finally filtering to eliminate false positives. This segmentation allows the system to maintain real-time processing speed while improving detection precision through systematic verification at each stage.

The patent introduces an intermediary verification mechanism between the initial heuristic detection and final curb identification. This intermediary step involves checking whether candidate points satisfy specific geometric conditions (constant first slope in plan view, constant sign second slope in profile view) and whether they form continuous curb structures. This intermediary verification acts as a filter that eliminates false detections while preserving true curb points, thereby improving measurement precision without sacrificing processing speed.

The patent applies partial action by implementing only the necessary verification steps required for accurate curb detection rather than using full complex algorithms like RANSAC. The system performs selective checks on candidate points (geometric constraint verification, continuity checks) that are sufficient to achieve high precision without the computational overhead of more complex algorithms. This partial verification approach maintains real-time processing capability while achieving accurate curb detection.

The patent changes the parameters of the detection algorithm by using simplified geometric constraints (constant slope conditions, constant sign conditions) instead of complex statistical methods. These parameter changes transform the detection problem into one that can be solved with simple comparisons and logical operations, enabling real-time processing while maintaining high precision. The approach replaces computationally intensive parameters with efficient geometric parameter checks.

The patent applies local quality by implementing different detection and verification strategies for different aspects of curb detection. The initial candidate detection uses simple heuristic methods, while the verification stage applies specific geometric constraints (constant slope in plan view, constant sign slope in profile view) locally to each candidate point. This localized verification approach maintains overall algorithm simplicity while improving precision through targeted checks where they are most needed.

The patent improves detection precision by adding dimensional analysis to the detection process. It examines candidate points not only in the horizontal plane but also in the vertical profile, checking for constant slope characteristics in both dimensions. This multi-dimensional verification approach enables the system to distinguish curbs from similar objects (walls, vehicles, uneven roads) by analyzing their geometric properties across multiple dimensions, thereby improving precision without significantly increasing algorithm complexity.