Vehicle Camera Extrinsic Calibration Using Sky Feature Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current camera calibration methods for autonomous vehicles are prone to errors due to long-term offset biases caused by slowly moving objects, such as clouds, which are misinterpreted as stationary objects, leading to inaccurate visual odometry and calibration drift.
Innovation Solution
A method that involves obtaining a sequence of images, determining feature points, identifying and removing feature points belonging to the sky and moving objects, and comparing camera-based motion trajectories with sensor-based trajectories to adjust extrinsic camera parameters, thereby improving calibration accuracy and reducing bias.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If visual odometry is used for camera calibration, then calibration can be performed dynamically, but long-term offset biases cause calibration drift and reduced accuracy
Solution Approach 1:
The patent extracts and removes feature points belonging to the sky from the set of feature points used in visual odometry. This is achieved by detecting sky regions in images and excluding feature points located within these regions, thereby eliminating the source of long-term offset biases while preserving the dynamic calibration capability
Solution Approach 2:
The patent implements a feedback mechanism where the calibration system continuously monitors feature points, identifies those belonging to the sky through sky region detection, and adjusts the feature point selection accordingly. This feedback loop prevents calibration drift by continuously correcting the input data for visual odometry
2Quantity of substance
If all detected feature points are used in visual odometry, then more data is available for calibration, but slowly moving objects like clouds introduce bias and reduce accuracy
Solution Approach 1:
The patent extracts feature points from the overall set and removes those located in sky regions. This selective extraction maintains a sufficient quantity of feature points for accurate visual odometry while eliminating the harmful subset (clouds and other moving sky objects) that would otherwise introduce bias
Solution Approach 2:
The patent applies different quality treatment to different regions of the image. Sky regions are identified and feature points within these regions are excluded, while feature points in non-sky regions are retained. This local quality approach ensures that only reliable feature points contribute to the motion estimation
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention is related to a computer-implemented method (100) for calibrating a set of extrinsic parameters of a camera mounted on a vehicle. The method (100) comprises: obtaining (S102) a sequence of images, captured by the camera, wherein each image of the sequence of images depicts a portion of a surrounding environment of the vehicle in a respective time instance of a plurality of time instances; determining (S104) a set of feature points in the sequence of images; determining (S106) an area representing a sky in the sequence of images; removing (S114), from the set of feature points, a subset of feature points, the subset of feature points belonging to the area representing the sky, thereby forming an updated set of feature points; determining (S118) a first motion trajectory of the vehicle over the plurality of time instances, based on the updated set of feature points; obtaining (S120) a second motion trajectory of the vehicle over the plurality of time instances, wherein the second motion trajectory is based on motion data obtained from one or more other sensors of the vehicle; and calibrating (S122) the camera by adjusting the set of extrinsic parameters of the camera based on a difference between the first motion trajectory and the second motion trajectory.