Histogram-Based L-Shape Detection in LiDAR Point Clouds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional L-shape clustering methods in LiDAR-based systems for object detection face high computational costs and poor performance when target vehicles lack well-defined corners or have objects extending from their sides, such as side-view mirrors.
Innovation Solution
A weighted histogram approach is employed to determine if a target object is L-shaped and to detect its pointing angle. This involves generating first and second weighted histograms for rotated target point clouds, with bins weighted by the number of points and distance spanned perpendicular to the bin width, to identify L-shape geometries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional L-shape clustering methods are used, then the detection process is simple, but the computational cost is high and performance is poor
Solution Approach 1:
The patent transforms the point cloud data by rotating it through multiple angles and generates histograms of point distributions. This parameter transformation converts the complex L-shape detection problem into a simpler histogram analysis problem, reducing computational complexity while improving detection reliability for vehicles with ambiguous corners or extended objects
Solution Approach 2:
The patent replaces traditional geometric clustering algorithms with a histogram-based statistical approach. Instead of mechanically searching for L-shaped configurations in point clouds, the system substitutes this with generating and analyzing histograms of rotated point distributions, which is computationally more efficient and more robust to variations in vehicle geometry
2Adaptability or versatility
If traditional L-shape clustering is used, then the method is straightforward, but it fails when vehicles lack well-defined corners or have extended objects
Solution Approach 1:
The patent makes the detection process dynamic by rotating the point cloud through multiple angles and analyzing histograms at each rotation. This dynamic approach allows the system to adapt to different vehicle orientations and geometries, improving robustness for vehicles with ambiguous corners or extended objects like side-view mirrors, while maintaining precise pointing angle measurement
Solution Approach 2:
The patent adds the dimension of rotation angles to the analysis. Instead of analyzing point clouds in a single fixed orientation, the system rotates the point cloud through multiple angles and analyzes the distribution of points in each rotated frame. This additional dimensional analysis enables the system to detect L-shapes regardless of their orientation and handles extended objects more effectively
Data Source
Figure 1a
Figure 1b
Figure 2
AI summary
A method of detecting L-shaped target objects within a point cloud includes rotating a target point cloud through a plurality of rotation angles and generating at least one weighted histogram for each of the plurality of rotation angles, wherein the weighted histogram includes a first plurality of bins, each bin having a width defined in a first axis, wherein each bin is weighted based on a number of points located within the bin and a distance between points in a direction perpendicular to a width of the bin. A score is generated for each of the plurality of rotation angles based on the at least one weighted histogram and determining whether a target point cloud is L-shaped based on the generated scores.