In-Vehicle LiDAR Downsampling for Robust Position Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for vehicle position estimation using lidar point cloud data down-sampling result in reduced accuracy and robustness when there are few objects around the vehicle, as the down-sampling process reduces the number of data points significantly.
Innovation Solution
An information processing device that adapts the size of down-sampling based on the number of associated measurement points, ensuring the number remains within a target range to maintain accuracy and robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If down-sampling process is applied to equalize dense point cloud data, then self position estimation accuracy is improved and calculation time is suppressed, but the number of point cloud data is greatly reduced and accuracy and robustness of estimation are lowered
Solution Approach 1:
The patent applies dynamics by making the down-sampling ratio adjustable rather than fixed. The control unit dynamically changes the down-sampling ratio based on the number of measurement points associated with each voxel, allowing the system to adapt to different environmental conditions and maintain optimal balance between accuracy and data quantity.
Solution Approach 2:
The patent changes the parameter of down-sampling ratio based on the association result between measurement points and voxels. When the number of associated measurement points is small, the system increases the down-sampling ratio to reduce data quantity; when the number is sufficient, it decreases the ratio to preserve accuracy, thus optimizing the balance between these two parameters.
2Loss of time
If down-sampling process is applied to reduce total number of point cloud data, then calculation time is suppressed, but the number of measurement points associated with each voxel is reduced and estimation robustness is lowered
Solution Approach 1:
The patent implements feedback by using the association result (number of measurement points per voxel) to determine the down-sampling ratio for the next iteration. This closed-loop control ensures that the system maintains sufficient measurement points for robust estimation while minimizing calculation time through appropriate data reduction.
Solution Approach 2:
The system dynamically adjusts the down-sampling ratio based on real-time association results, allowing it to adapt to varying environmental conditions and maintain optimal balance between processing speed and estimation reliability in different scenarios.
3Device complexity
If fixed down-sampling ratio is used, then processing is simplified, but the system cannot adapt to different environments with varying numbers of objects and measurement points
Solution Approach 1:
The patent transforms the static fixed ratio system into a dynamic adaptive system. The control unit automatically adjusts the down-sampling ratio based on the number of measurement points associated with each voxel, enabling the system to adapt to different environmental conditions without manual intervention.
Solution Approach 2:
The system performs self-adjustment by automatically determining the appropriate down-sampling ratio based on the association result between measurement points and voxels. This self-service mechanism eliminates the need for external configuration and allows the system to optimize itself for different environments.
Data Source
AI summary
The controller 13 of the in-vehicle device 1 acquires point cloud data outputted by a lidar 2. Then, the controller 13 generates processed point cloud data obtained by down-sampling the point cloud data. The controller 13 matches the processed point cloud data with voxel data VD which represents a position of an object with respect to each voxel that is a unit area and thereby associates a measurement point of the processed point cloud data with the each voxel. Then, the controller 13 changes the size of the subsequent down-sampling based on the associated measurement points number Nc which is the number of measurement points associated with the each voxel among the measurement points of the processed point cloud data.