Camera Pose Refinement via Adaptive Bundle Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Bundle adjustment in Structure from Motion and Multi-View-Stereo 3D reconstruction is bottlenecked by noisy camera parameters from onboard sensors, which require refinement for downstream processing, and existing methods rely on initial filtering and outlier elimination techniques like RANSAC, limiting efficiency and robustness.
Innovation Solution
A novel BA pipeline that directly uses approximate camera parameters from low-precision sensors and unfiltered putative matches, incorporating a persistency factor-based adaptive robust function to weight residuals and optimize camera pose refinement without explicit outlier filtering, leveraging camera metadata and feature tracks for efficient refinement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If RANSAC-based model estimation algorithm is used to obtain initial camera poses, then camera pose estimation can be performed, but the process is time-consuming and requires early-stage filtering to eliminate outliers
Solution Approach 1:
The patent uses GPS and IMU sensors to obtain preliminary camera pose estimates before bundle adjustment. These sensor-based preliminary estimates serve as high-quality initial values for the BA optimization, eliminating the need for time-consuming RANSAC-based model estimation and early-stage filtering. The sensors provide ready-to-use position and orientation data that can be directly integrated into the BA pipeline.
2Measurement precision
If GPS and IMU measurements are used as complementary values with RANSAC or EKF filtering, then camera parameters can be refined, but the process becomes complex and requires multiple processing stages
Solution Approach 1:
The patent merges GPS position data and IMU orientation data directly into the bundle adjustment optimization framework. Instead of using separate filtering stages (RANSAC, EKF) followed by BA, the sensor measurements are integrated as additional constraints and observations within the unified BA formulation. This combining approach simplifies the pipeline while maintaining refinement accuracy.
Solution Approach 2:
The patent extracts and removes the complex intermediate filtering stages (RANSAC, EKF) from the traditional pipeline. By directly using sensor measurements as initial values and constraints in BA, the method eliminates the need for multiple sequential processing stages, reducing overall pipeline complexity while preserving the essential function of noise reduction through the BA optimization itself.
3Reliability
If traditional BA pipeline with RANSAC filtering is used, then outliers can be eliminated, but the computational efficiency is reduced and robustness to noisy sensor data is limited
Solution Approach 1:
The patent changes the approach from discrete outlier elimination (RANSAC iteratively removing outliers) to continuous robust optimization (BA with robust cost functions). The bundle adjustment framework inherently handles noisy sensor data through its optimization formulation, which minimizes reprojection errors across all observations. This parameter change from filtering to optimization maintains reliability while improving computational efficiency by processing all data points simultaneously in a unified framework.
Data Source
AI summary
The present invention provides a method for fast, robust and efficient BA pipeline (SfM) for wide area motion imagery (WAMI). The invention can, without applying direct outliers filtering (e.g. RANSAC) or re-estimation of the camera parameters (e.g. essential matrix estimation) efficiently refine noisy camera parameters in very short amounts of time. The method is highly robust owing to its adaptivity with the persistency factor of each track. The present invention highly suitable for sequential aerial imagery, particularly for WAMI, where camera parameters are available from onboard sensors.