LiDAR Point Cloud Segmentation Using Sphere Projection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
LiDAR-based vehicle tracking systems face challenges such as limited observation of vehicles far away, occlusion leading to fractured point cloud responses, and high processing power requirements in driver assistance systems or autonomous vehicles.
Innovation Solution
A method for vehicle tracking using LiDAR point cloud data that projects points to a unit sphere, forms clusters based on Euclidean distance, and utilizes tracked vehicle representation from previous frames to aid clustering, thereby accelerating computing and improving segmentation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional LiDAR point cloud processing is used, then vehicle tracking is performed, but processing power requirements are high and segmentation accuracy decreases for distant or occluded vehicles
Solution Approach 1:
The patent applies segmentation by dividing the point cloud into multiple clusters based on spatial proximity. Each point is assigned to a cluster if it falls within a threshold distance from cluster center points, enabling efficient processing of large point clouds while maintaining segmentation accuracy for vehicle identification.
Solution Approach 2:
The patent transforms the 3D point cloud data into a 2D projected image space. By projecting points onto an image plane and processing in this transformed dimension, the system reduces computational complexity while preserving essential spatial relationships needed for accurate vehicle segmentation.
2Productivity
If traditional clustering methods are used, then point cloud processing is performed, but computing speed is slow for real-time vehicle tracking
Solution Approach 1:
The patent performs preliminary actions by first projecting all points to the image plane and identifying cluster center points before assigning remaining points to clusters. This pre-processing step organizes the data structure to enable faster subsequent clustering operations while maintaining segmentation accuracy.
Solution Approach 2:
The clustering algorithm uses self-service by allowing points to automatically assign themselves to the nearest cluster center based on threshold distance criteria. This eliminates the need for complex iterative optimization algorithms, significantly improving computing speed while maintaining reliable segmentation.
3Reliability
If complete vehicle observation is required, then all vehicle parts must be detected, but distant vehicles are poorly observed and occlusion causes fractured point cloud responses
Solution Approach 1:
The patent merges fragmented point cloud clusters that belong to the same vehicle by comparing spatial relationships and vehicle attributes across multiple frames. Clusters representing different parts of the same occluded vehicle are combined into a single coherent vehicle detection, maintaining detection completeness despite partial occlusion.
Solution Approach 2:
The system uses feedback from previously tracked vehicle representations to guide current frame clustering. By incorporating historical vehicle position, size, and shape information, the algorithm can correctly assign ambiguous points and maintain reliable vehicle detection even when current observations are incomplete or occluded.
Data Source
AI summary
A method of LiDAR-based vehicle tracking for a non-transitory computer readable storage medium storing one or more programs is disclosed. The one or more programs include instructions, which when executed by a computing device, cause the computing device to perform the following steps comprising: projecting points in a processed point cloud to a unit sphere, for each of the projected points, determining that a distance between one and each of its neighboring points is smaller than a threshold, and forming a cluster of the one and its neighboring points.