Stereo Visual Odometry Pose Estimation Drift Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing stereo-based visual odometry systems face challenges such as drift due to errors in calibration, image quality, feature position inaccuracies, and instability in motion estimation, particularly in environments with moving objects or non-texture surfaces, and lack a metric to stop tracking when feature points become insufficient for pose estimation.
Innovation Solution
A method and system for estimating pose from a sequence of images by detecting and matching Harris corner feature points in stereo images, tracking these points across frames, and recalculating pose based on three-dimensional coordinates, with constraints to reduce false matches and dynamically updating feature points to maintain accurate pose estimation, using a spatial distribution metric to determine when to re-initialize tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If frame-by-frame pose estimation is used in stereo-based visual odometry, then computational simplicity is maintained, but navigation error increases significantly over time due to drift accumulation
Solution Approach 1:
The system performs preliminary 3D reconstruction of feature points from stereo image pairs before pose estimation. By pre-computing accurate 3D positions of feature points using triangulation, the system establishes a reliable spatial reference framework that reduces drift accumulation in subsequent pose estimations across multiple frames.
Solution Approach 2:
The system maintains continuous tracking of feature points across multiple video frames, updating pose estimates incrementally while preserving the continuity of spatial relationships. This continuous tracking approach prevents drift accumulation by consistently referencing the established 3D feature point positions throughout the video sequence.
2Measurement precision
If multi-frame tracking is performed to reduce navigation error, then pose estimation accuracy improves, but the system lacks a metric to determine when tracking should stop, leading to potential error propagation
Solution Approach 1:
The system computes pose estimation error metrics after each frame's pose calculation and uses this feedback to determine whether to continue or terminate tracking. By monitoring the accuracy of pose estimates and the quality of feature point matches, the system can objectively decide when tracking should stop, preventing error propagation while maintaining high accuracy.
Solution Approach 2:
The system replaces manual or heuristic tracking management with an automated error-metric-driven decision system. By substituting complex tracking management logic with objective error metrics and automated thresholds, the system achieves accurate tracking termination without requiring complex manual intervention or heuristic rules.
3Measurement precision
If stereo matching is performed without geometric constraints to reduce false matches, then matching accuracy improves, but computational time increases significantly
Solution Approach 1:
The system performs preliminary stereo calibration to establish geometric constraints (epipolar geometry, fundamental matrix) before stereo matching. By pre-computing these constraints, the system can efficiently restrict the search space for feature point matches to epipolar lines, significantly reducing computational time while maintaining high matching accuracy through geometric validation.
Solution Approach 2:
The stereo matching process is segmented into distinct stages: feature detection, epipolar constraint application, candidate match generation, and geometric validation. By dividing the matching process into these segments, the system applies geometric constraints at optimal points to eliminate false matches efficiently without requiring exhaustive search of all possible correspondences.
4Quantity of substance
If feature points from moving objects are included in pose estimation, then feature point quantity increases, but pose accuracy deteriorates due to incorrect correspondences
Solution Approach 1:
The system computes residuals between observed and expected feature point positions based on the estimated pose and uses this feedback to identify and reject outliers. By monitoring the consistency of feature point correspondences with the computed pose, the system can detect and eliminate feature points from moving objects that would otherwise corrupt the pose estimation, maintaining accuracy while utilizing abundant feature points.
Solution Approach 2:
The system dynamically adjusts the reliability weight or inclusion criteria for feature points based on their consistency with the estimated pose and spatial distribution patterns. By changing the parameters of feature point selection and weighting based on observed correspondence quality, the system can utilize feature points from stationary objects for accurate pose estimation while automatically downweighting or excluding points from moving objects.
Data Source
AI summary
A method for estimating pose from a sequence of images, which includes the steps of detecting at least three feature points in both the left image and right image of a first pair of stereo images at a first point in time; matching the at least three feature points in the left image to the at least three feature points in the right image to obtain at least three two-dimensional feature correspondences; calculating the three-dimensional coordinates of the at least three two-dimensional feature correspondences to obtain at least three three-dimensional reference feature points; tracking the at least three feature points in one of the left image and right image of a second pair of stereo images at a second point in time different from the first point in time to obtain at least three two-dimensional reference feature points; and calculating a pose based on the at least three three-dimensional reference feature points and its corresponding two-dimensional reference feature points in the stereo images. The pose is found by minimizing projection residuals of a set of three-dimensional reference feature points in an image plane.