Camera Pose Estimation via Lane Marker Map Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing camera pose estimation methods for autonomous vehicles are inadequate in accurately determining the orientation and position of cameras attached to vehicles due to vibrations and non-rigid mechanical structures, which affects precise object detection and distance estimation.
Innovation Solution
A method that estimates camera pose by matching 2D pixel locations of lane markers from images with 3D world coordinates stored in a high-definition map, using deep learning lane detection and minimizing the sum of squared distances, while incorporating a constraint term to limit parameter search space and account for previous frame poses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If camera pose estimation is performed using traditional methods, then the system is simple to implement, but the measurement precision of camera orientation and position deteriorates due to vibrations and non-rigid mechanical structures
Solution Approach 1:
The patent introduces lane markers as an intermediary reference object between the camera and the road. By detecting lane markers in images and matching them with pre-stored map data, the system creates a reliable reference framework that mediates the pose estimation process, overcoming the instability caused by vehicle vibrations and non-rigid mounting structures
Solution Approach 2:
The patent replaces traditional mechanical pose estimation methods (which rely on physical sensors and rigid mounting) with a vision-based computational approach. Instead of using mechanical gyroscopes or accelerometers, the system uses image processing and optimization algorithms to estimate camera pose, thereby eliminating the limitations of mechanical systems under vibration
2Measurement precision
If the parameter search space is not constrained, then the optimization can find the global minimum, but the computation time increases significantly affecting real-time performance
Solution Approach 1:
The patent performs preliminary actions by pre-storing detailed lane marker information in high-definition maps before the vehicle reaches those locations. During real-time operation, the system only needs to match detected lane markers against this pre-prepared data, significantly reducing computation time while maintaining accurate pose estimation
Solution Approach 2:
The patent applies local quality by constraining the optimization search to local neighborhoods around initial pose estimates rather than searching the entire parameter space. This localized optimization approach maintains precision by focusing on relevant regions while achieving real-time performance through reduced computational scope
3Adaptability or versatility
If lane markers are detected from images, then the system can estimate camera pose, but the measurement precision deteriorates when lane markers are obscured or poorly visible
Solution Approach 1:
The patent merges multiple sources of information including detected lane markers, pre-stored HD map data, and optimization constraints to estimate camera pose. This combination allows the system to maintain adaptability by using available lane markers while improving precision through the complementary information from multiple sources, especially when some lane markers are obscured
Data Source
AI summary
Techniques are described for estimating pose of a camera located on a vehicle. An exemplary method of estimating camera pose includes obtaining, from a camera located on a vehicle, an image including a lane marker on a road on which the vehicle is driven, and estimating a pose of the camera such that the pose of the camera provides a best match according to a criterion between a first position of the lane marker determined from the image and a second position of the lane marker determined from a stored map of the road.