3D Data Discriminative Recovery via Background Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for reconstructing three-dimensional surface information often include unwanted background data, requiring manual or automatic processes to isolate the target of interest, which can be tedious and error-prone, especially in complex environments, and fail to provide scaled 3D data.
Innovation Solution
A method and system using computer vision to selectively reconstruct 3D data from a target of interest by filtering out background data through video processing, employing algorithms like GrabCut for segmentation and optical ray tracing to generate accurate, scaled 3D models without manual intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional 3D reconstruction methods (SFM, structured light, TOF) are used, then 3D surface information can be obtained, but unwanted background data is always reconstructed along with the target of interest
Solution Approach 1:
The patent segments the 3D point cloud data into target object and background components by analyzing surface normal vectors and spatial distribution characteristics. This segmentation allows selective retention of target data while removing background data, resolving the contradiction between obtaining complete 3D information and avoiding background contamination.
Solution Approach 2:
The patent extracts the target object from the reconstructed 3D data by identifying and isolating points that belong to the target based on geometric and spatial criteria. This extraction process removes unwanted background information while preserving the target's 3D surface information, directly addressing the contradiction.
2Loss of information
If manual processes (digital tool cut) are used to remove unwanted background, then 3D data isolation can be achieved, but the process becomes very tedious
Solution Approach 1:
The patent implements self-service automation where the system automatically performs background removal through algorithmic analysis of the 3D point cloud data. The system uses surface normal vector analysis and spatial distribution algorithms to autonomously identify and separate target objects from background, eliminating the need for tedious manual digital cutting while maintaining accurate 3D data isolation.
3Productivity
If automatic 3D algorithms are used to remove unwanted background, then processing efficiency is improved, but the algorithms often fail for complex background or complex 3D surface boundaries
Solution Approach 1:
The patent applies local quality analysis by examining surface normal vectors and spatial relationships at local regions of the 3D point cloud. This localized analysis allows the algorithm to adapt to complex boundaries and varying geometric characteristics differentially across the data, improving reliability for complex backgrounds while maintaining automated processing efficiency.
Solution Approach 2:
The patent changes parameters such as surface normal vector thresholds and spatial distribution criteria to adapt to different complexity levels of backgrounds and boundaries. By dynamically adjusting these parameters, the algorithm maintains high reliability across varied scenarios while preserving automated processing efficiency.
4Measurement precision
If SFM based approach is used for 3D reconstruction, then 3D data can be recovered, but the recovered 3D data is unscaled
Solution Approach 1:
The patent introduces an intermediary scaling process that uses reference objects or known geometric constraints within the captured images to establish scale relationships. This intermediary step converts the unscaled SFM output into scaled 3D data by applying transformation matrices derived from reference measurements, thereby achieving both 3D data recovery and scale accuracy.
Data Source
AI summary
A method of discriminative recovery of three-dimensional digital data of a target of interest in a cluttered or controlled environment. The method uses the traditional structure from motion (SFM) 3D reconstruction method on a video of a target of interest to track and extract 3D sparse feature points and relative orientations of the video frames. Desired 3D points are filtered from the 3D sparse feature points in accordance to a user input and a digital cutting tool. Segmented images are generated by separating the target of interest from the background scene in accordance to the desired 3D points. A dense 3D reconstruction of the target of interest is generated by inputting the segmented images and the relative orientations of the video frames.