3D Scanning Pose Refinement with Geometry–Texture Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional 3D scanning methods suffer from poor adaptability to optical changes, unstable registration results, large pose estimation deviations, low optimization efficiency, and ineffective mechanisms for filtering false matching results, leading to inefficient data processing.
Innovation Solution
An online matching and optimization method combining geometry and texture, utilizing a 3D scanning device with a depth sensor and camera, estimates a preliminary pose, refines it using geometric and texture information, and employs multi-mode pose optimization with segmentation to eliminate errors, incorporating a penalty factor for robust and accurate registration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional 3D scanning methods are used, then the scanning process is simple, but the adaptability to optical changes is poor and registration results are unstable
Solution Approach 1:
The patent combines geometric constraints and texture constraints to form a composite optimization framework. This composite approach integrates multiple types of information (depth geometry and texture appearance) to simultaneously improve adaptability to optical changes and stability of registration results, resolving the contradiction between these two parameters.
Solution Approach 2:
The patent dynamically adjusts optimization parameters and penalty factors based on the characteristics of different data segments. By changing parameters adaptively according to optical conditions and data quality, the system maintains both high adaptability to optical changes and stable registration results.
2Productivity
If all image frames are optimized together, then comprehensive optimization is achieved, but the optimization efficiency is low due to large data processing amount
Solution Approach 1:
The patent divides the optimization process into multiple stages: initial pose estimation, segmented optimization of key frames, and global optimization. This segmentation reduces the computational burden of optimizing all frames simultaneously while maintaining pose estimation accuracy through multi-stage refinement.
Solution Approach 2:
The patent performs preliminary pose estimation and key frame selection before the main optimization process. This preliminary action pre-processes the data to identify important frames and initial poses, enabling more efficient subsequent optimization with reduced computational requirements while preserving accuracy.
3Reliability
If conventional matching methods are used, then the process is straightforward, but false matching results cannot be effectively filtered
Solution Approach 1:
The patent implements a penalty mechanism that provides feedback during the optimization process. By introducing penalty terms for geometric and texture constraints, the system evaluates and filters matching results, rewarding consistent matches and penalizing outliers, thereby improving matching accuracy through iterative feedback.
Solution Approach 2:
The patent incorporates penalty factors in advance to prevent false matches. By pre-defining geometric and texture consistency constraints with penalty terms, the optimization process actively prevents erroneous matches from being accepted, rather than merely filtering them after the fact.
Data Source
AI summary
An online matching and optimization method combining geometry and texture and a three-dimensional (3D) scanning system are provided. The method includes obtaining pairs of depth texture images with a one-to-one corresponding relationship, and collecting the pairs of the depth texture images including depth images by a depth sensor and collecting texture images by a camera device; adopting a strategy of coarse to fine to perform feature, matching on the depth texture images corresponding to a current frame and on the depth texture images corresponding to the target frames, to estimate a preliminary pose of the depth sensor in the 3D scanning system; combining a geometric constraint and a texture constraint to optimize the estimated preliminary pose, and obtaining a refined motion estimation between the frames.