Sparse Image Point Correspondences Refinement for Autonomous Vehicle Ground Truth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating ground truth datasets for autonomous vehicles are costly and inefficient, with existing datasets often not well generalized across different environments, making it challenging to develop and validate algorithms for autonomous driving systems.
Innovation Solution
A system and method for generating a ground truth dataset for motion planning, involving sensor calibration and time synchronization, which includes a computing module to compute camera poses using GNSS-insertion estimates, and modules for generating and refining sparse image point correspondences to create a calibrated LiDAR static-scene point cloud, enabling accurate data alignment and storage for motion planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional methods are used to generate ground truth datasets for autonomous vehicles, then the datasets can be created with established procedures, but the cost of processing and analyzing data becomes excessively high and efficiency is reduced
Solution Approach 1:
The system segments the ground truth generation process into distinct modules: LiDAR static scene point cloud generation, image point correspondence generation, and refinement processes. Each module handles specific tasks independently, improving processing efficiency while maintaining accuracy through specialized optimization in each segment.
Solution Approach 2:
The system performs preliminary actions by pre-processing LiDAR data to create static scene point clouds and pre-establishing correspondence relationships between images and 3D points before final dataset generation. This preliminary structuring reduces computational burden during actual data processing and improves overall efficiency.
2Quantity of substance
If existing datasets are used for autonomous vehicle development, then data availability is maintained, but the datasets do not generalize well across different environments
Solution Approach 1:
The system generates ground truth data that is universally applicable across different environments and sensor configurations. By creating correspondence relationships between LiDAR point clouds and images using geometric projections that are environment-agnostic, the generated datasets can be generalized to various autonomous driving scenarios and sensor setups.
Solution Approach 2:
The system varies parameters such as camera poses, LiDAR configurations, and scene geometries to generate diverse ground truth datasets that adapt to different environmental conditions. This parameter variation ensures the datasets generalize well across multiple environments while maintaining quantitative rigor.
3Measurement precision
If detailed sensor calibration and time synchronization procedures are implemented, then data alignment accuracy is improved, but the complexity of the system increases
Solution Approach 1:
The system introduces an intermediary LiDAR static scene point cloud as a mediator between multiple sensors. This intermediary representation provides a common reference frame that simplifies calibration and time synchronization procedures, reducing system complexity while maintaining high alignment accuracy through the mediating geometric model.
Data Source
AI summary
A system for generating a ground truth dataset for motion planning of a vehicle is disclosed. The system includes an internet server that further includes an I/O port, configured to transmit and receive electrical signals to and from a client device; a memory; one or more processing units; and one or more programs stored in the memory and configured for execution by the one or more processing units, the one or more programs including instructions for: a corresponding module configured to correspond, for each pair of images, a first image of the pair to a LiDAR static-scene point cloud; and a computing module configured to compute a camera pose associated with the pair of images in the coordinate of the point cloud.