LiDAR Grid-Map Tracking for Accurate Elongated Object Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of clusters formed by clustering point clouds acquired through LiDAR sensors in autonomous vehicles is low, necessitating improved methods for object tracking.
Innovation Solution
An apparatus and method using a LiDAR sensor that generates a grid map and applies static and dynamic labeling techniques based on threshold values and angular differences to enhance clustering accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fixed threshold value is used for clustering point clouds, then the clustering process is simple and fast, but the clustering accuracy is low
Solution Approach 1:
The patent applies dynamics by transforming the clustering algorithm from a static fixed-threshold approach to a dynamic adaptive-threshold approach. The threshold values are no longer constant but are dynamically adjusted based on the local characteristics of point clouds, including spatial distribution density, angular relationships between adjacent points, and distance from the LiDAR sensor. This dynamic adaptation allows the clustering algorithm to automatically optimize its parameters for different regions and objects, significantly improving clustering accuracy while maintaining computational efficiency through localized rather than globally complex processing.
Solution Approach 2:
The patent implements local quality by applying different threshold values for different spatial regions and object types. Instead of using a uniform threshold across the entire point cloud, the system calculates and applies localized thresholds based on regional characteristics such as point density, angular separation, and distance from the sensor. This allows the clustering algorithm to adapt to local variations in object geometry and point distribution, improving the precision of object boundaries and internal structure identification in different areas of the scene.
2Measurement precision
If simple clustering is used, then processing speed is fast, but object detection precision is insufficient
Solution Approach 1:
The patent applies segmentation by dividing the point cloud processing into distinct stages: initial rapid clustering using a base threshold, followed by refined clustering in regions requiring higher precision. The algorithm segments the processing based on point density thresholds, angular relationship criteria, and distance-based regions of interest. This multi-stage segmentation allows the system to perform fast processing for simple regions while applying more computationally intensive precision clustering only where necessary, thereby improving overall object detection precision without proportionally increasing total processing time.
3Measurement precision
If dynamic threshold values are used for clustering, then clustering accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent implements partial action by applying dynamic threshold adjustment selectively rather than uniformly across all point clouds. The system identifies regions requiring enhanced clustering precision based on criteria such as point density variations, angular relationships, and distance from the LiDAR, and applies computationally intensive dynamic thresholding only in those specific regions. In simpler regions, the system uses faster fixed-threshold or coarser dynamic adjustment methods. This partial application of complex processing significantly reduces the overall computational power required while maintaining high clustering accuracy in critical areas.
Data Source
AI summary
An object-tracking method using a LiDAR sensor according to an embodiment includes: generating a grid map based on a point cloud generated with respect to an object using a LiDAR sensor mounted to a vehicle traveling in an x-axis direction; performing static labeling when the absolute value of the y-coordinate of the center of each grid on the grid map in a y-axis direction that intersects the x-axis direction is less than a y-axis coordinate threshold value; and performing dynamic labeling when the absolute value is not less than the y-axis coordinate threshold value. Since both static labeling and dynamic labeling are utilized, it is possible to improve the performance of outputting information about an object that is elongated, thus improving the performance of an autonomous driving system, such as prediction of a travel route and map matching.