LiDAR Feature Tracking Using 3D Point Cloud Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for tracking features of objects using sensor data are not sufficiently effective, particularly in real-time applications such as driver assistance and autonomous driving, where improved accuracy and handling of occlusion situations are needed.
Innovation Solution
A method implemented by a computer that processes data from a LiDAR sensor by obtaining two matrices of points representing the environment at different instants, projecting these into two-dimensional images, determining features using 3D coordinates and intensity values, and comparing windows of neighboring points to track feature displacement over time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature tracking is performed using conventional image processing methods, then processing speed is maintained, but tracking accuracy deteriorates in occlusion situations
Solution Approach 1:
The patent transitions from two-dimensional image plane processing to three-dimensional point cloud processing by incorporating depth information from LiDAR. Features are tracked in 3D space using spatial coordinates (x, y, z) and intensity values, enabling accurate tracking even when features are occluded in the 2D image plane. The second window of neighboring points is defined in 3D space with a depth threshold, allowing the system to distinguish between overlapping features at different depths.
2Measurement precision
If 3D point cloud data is processed directly, then tracking accuracy in occlusion situations is improved, but processing time increases
Solution Approach 1:
The patent segments the point cloud data by defining localized windows of neighboring points around each feature. Instead of processing the entire point cloud, the method focuses computational resources on small regions (first window in 2D projection, second window in 3D space) containing the feature and its immediate neighbors. This segmentation dramatically reduces processing time while maintaining tracking accuracy.
Solution Approach 2:
The patent applies partial action by processing only the necessary subset of data (points within the second window of neighboring points) rather than the complete point cloud. The depth threshold parameter controls the extent of processing, including only points that are sufficiently close in 3D space, thus reducing computational load while sufficient for accurate tracking.
3Measurement precision
If intensity values are utilized in feature determination, then tracking precision is improved, but computational complexity increases
Solution Approach 1:
The patent merges multiple data types (spatial coordinates x, y, z and intensity values) into a unified feature description. The feature is characterized by combining its 3D position with its intensity property, creating a more robust descriptor that remains valid under occlusion. The comparison of second windows involves both spatial proximity and intensity similarity, providing redundant information that improves tracking precision without requiring complex separate processing streams.
Data Source
AI summary
Examples set out a method of processing data acquired by a LiDAR sensor, a computer configured to carry out this processing, a vehicle carrying a computer thus configured, a computer program product, and a computer-readable non-transitory storage medium.