Autonomous Vehicle Sensor Registration via Pose-Based Trajectory Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing calibration methods for cameras and positioning devices in autonomous vehicles suffer from inaccurate trajectory alignment due to positioning dimension degradation, scale ambiguity, and asynchronous data acquisition, leading to poor registration and calibration effects.
Innovation Solution
A registration method and apparatus that generate and align trajectory curves based on detection poses of vehicle-mounted cameras and positioning devices at multiple time points, using curve fitting and alignment techniques to improve accuracy and registration precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If time-based alignment is used for trajectory alignment, then the alignment process is simple, but positioning accuracy degrades and registration precision is poor
Solution Approach 1:
The patent changes the alignment parameter from time-based to pose-based. Instead of aligning trajectories using time stamps, the system uses geometric pose information (position and orientation) of the vehicle at different locations. This parameter change resolves the contradiction by maintaining operational simplicity through pose matching while significantly improving registration precision through geometric consistency.
2Adaptability or versatility
If asynchronous data acquisition is used, then data collection flexibility is improved, but trajectory alignment accuracy deteriorates due to scale ambiguity
Solution Approach 1:
The patent introduces pose information as an intermediary to resolve the scale ambiguity problem in asynchronous data acquisition. By using the vehicle's pose (position and orientation) at different locations as a common reference frame, the system can align trajectories from asynchronously acquired camera and positioning device data without suffering from scale ambiguity, thus maintaining both data collection flexibility and alignment accuracy.
3Ease of operation
If existing calibration methods are used, then the calibration process is straightforward, but positioning dimension degradation occurs leading to poor calibration results
Solution Approach 1:
The patent replaces the traditional time-based mechanical alignment system with a pose-based geometric alignment system. Instead of relying on time-synchronized data acquisition and mechanical coordinate transformation, the system uses geometric pose matching to establish coordinate relationships. This substitution eliminates positioning dimension degradation while maintaining calibration process straightforwardness through pose-based trajectory alignment.
Data Source
AI summary
A registration method and a registration apparatus for an autonomous vehicle is provided. The method includes: generating a first trajectory curve based on first detection poses of a vehicle-mounted camera at a plurality of first time points, generating a second trajectory curve based on second detection poses of a vehicle-mounted positioning device at a plurality of second time points, aligning the first trajectory curve with the second trajectory curve based on the first detection poses and the second detection poses, and registering the vehicle-mounted camera and the vehicle-mounted positioning device based on the first trajectory curve and the second trajectory curve aligned with each other.