Floorplan-Aware Camera Pose Refinement for Indoor 3D Scan Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for refining camera poses in large indoor scenes are error-prone, leading to distorted geometry and inaccuracies in 3D scans due to accumulation of small pose errors and issues with keypoint matching, especially in scenes with repetitive patterns and textureless surfaces.
Innovation Solution
A method and system for floorplan-aware camera pose refinement that incorporates prior knowledge of the scene structure from a floorplan, using a three-term cost function that includes geometric, floor, and wall terms to optimize camera poses, aligning structural elements of the scan with the floorplan.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional bundle adjustment is used to refine camera poses, then computational efficiency is maintained, but geometric accuracy and structural consistency deteriorate due to error accumulation
Solution Approach 1:
The patent applies preliminary action by incorporating floorplan constraints before final pose optimization. The method first extracts structural elements (walls, floors) from the floorplan and uses them as prior knowledge to guide the bundle adjustment process, preventing error accumulation before it occurs rather than correcting it afterward.
Solution Approach 2:
The patent uses floorplan structural elements as an intermediary between raw camera poses and final refined poses. The walls and floors extracted from the floorplan act as a mediator that constrains and guides the optimization process, ensuring geometric consistency while maintaining computational efficiency.
2Area of stationary object
If more scans are taken to cover large scenes, then scene coverage improves, but pose error accumulation increases leading to geometry corruption
Solution Approach 1:
The patent implements feedback by continuously comparing reconstructed geometric elements (walls, floors) against the ground truth floorplan during the optimization process. This feedback loop allows the system to correct pose errors incrementally across multiple scans, maintaining reliability even as scene coverage expands.
Solution Approach 2:
The patent segments the large scene into manageable structural components (individual walls, floors, and their corresponding floorplan elements). By optimizing each segment independently with floorplan constraints and then integrating them, the system maintains scan consistency across the entire large scene without error accumulation.
3Productivity
If standard keypoint matching is used in textureless environments, then processing speed is maintained, but matching accuracy deteriorates due to similar patterns
Solution Approach 1:
The patent uses floorplan-derived structural elements as an intermediary to replace unreliable keypoint matching in textureless environments. Instead of matching keypoints on repetitive or textureless surfaces, the system matches reconstructed geometric primitives (walls, floors) against the floorplan, achieving both speed and accuracy.
Solution Approach 2:
The patent substitutes the mechanical keypoint matching system with a geometric constraint-based system. Rather than relying on image-based feature detection and matching, the system uses geometric relationships derived from the floorplan to directly constrain camera poses, eliminating the accuracy-speed tradeoff in textureless environments.
Data Source
Figure 1
Figure 2a~2b
Figure 3
AI summary
The invention relates to the field of computer vision. A method for floorplan-aware camera pose refinement is provided, the method comprising receiving a floorplan of three-dimensional structure of a scene which reflects the prior structure of the scene; receiving one or more imaging scans of the scene; identifying a gravity direction for the one or more imaging scans; constructing a boundary scan by filtering out points in the one or more imaging scans whose horizontal projections are statistical outliers; aligning the boundary scan with the floorplan; extracting structural elements in the one or more imaging scans; and optimizing a three-term cost function including a geometric term, a floor term and a walls term by pulling points in the one or more imaging scans corresponding to structural elements, to respective points in the floorplan.