LiDAR Object Recognition via Contour Point Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for recognizing object shapes using point clouds from lidar sensors, such as the convex hull algorithm, fail to accurately identify objects, especially those with non-convex shapes near autonomous vehicles, leading to inefficiencies in obstacle detection and driving stability.
Innovation Solution
An apparatus and method that extracts valid contour points from point clouds using a controller to generate a contour line, involving sample points, reference angles, and filtering processes to accurately recognize object shapes, improving the accuracy of object recognition on roads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a convex hull algorithm is used to recognize object shapes from point clouds, then the processing is simple and fast, but the recognition accuracy is poor for non-convex shapes
Solution Approach 1:
The patent divides the point cloud data into multiple clusters based on spatial proximity and characteristics, then processes each cluster separately to extract contour points. This segmentation approach enables accurate recognition of non-convex shapes by handling each region independently, while maintaining computational efficiency through localized processing.
Solution Approach 2:
The patent transitions from 3D point cloud data to 2D contour representation by extracting valid contour points that form closed loops. This dimensional reduction from three-dimensional points to two-dimensional contour lines simplifies the shape recognition problem while preserving essential geometric information for accurate object identification.
2Measurement precision
If all points in the point cloud are processed to extract contour points, then the recognition accuracy is high, but the processing time and computational load increase
Solution Approach 1:
The patent extracts only the essential valid contour points that define object boundaries, discarding redundant interior points. By identifying and extracting only the points that form the outer contour and significant internal features, the system achieves accurate shape recognition with reduced computational load compared to processing all point cloud data.
Solution Approach 2:
The patent performs preliminary clustering and filtering of point cloud data before contour extraction, organizing points into meaningful groups based on spatial relationships. This preliminary processing step reduces the data volume and identifies candidate contour points in advance, significantly reducing the computational complexity of the subsequent contour extraction process.
Data Source
AI summary
An apparatus for recognizing an object based on a lidar sensor and a method thereof can extract a valid contour point of an object from a point cloud generated by the lidar sensor and recognize a shape of the object based on a contour line composed of the extracted valid contour point of the object, such that it is possible to recognize the shape of an object located on a road with high accuracy, thereby improving the driving stability of an autonomous vehicle. The apparatus includes the lidar sensor that generates the point cloud around an autonomous vehicle, and a controller that extracts the valid contour point of the object from the point cloud and recognizes the shape of the object based on the contour line composed of the extracted valid contour point.