TSDF Volumetric 3D Scanning for Low-Complexity Mobile Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D scanning methods struggle to efficiently combine depth-camera image information to create a comprehensive representation of objects or environments, particularly on resource-constrained mobile devices operating in unconstrained environments.
Innovation Solution
A method and apparatus using a truncated signed distance function (TSDF) to fuse depth maps into a volumetric representation, employing pose estimation techniques like iterative closest point (ICP), volumetric gradient descent (VGD), and structure-from-motion (SfM) to generate a stable and efficient 3D model, avoiding the need for special auxiliary data structures and relying on fixed-point math.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional 3D scanning methods are used to combine depth-camera image information, then a 3D model can be generated, but the computational complexity is high and it is not suitable for resource-constrained mobile devices
Solution Approach 1:
The patent changes the mathematical representation parameters by using a volumetric grid system with fixed-point arithmetic instead of traditional floating-point calculations. This parameter change reduces computational complexity while maintaining 3D reconstruction accuracy, making the system suitable for resource-constrained mobile devices.
Solution Approach 2:
The patent replaces complex mechanical computation systems with a simplified volumetric approach using fixed-point math. By substituting traditional computational mechanics with a grid-based volumetric system, the patent reduces processing requirements while maintaining scanning efficiency.
2Adaptability or versatility
If existing 3D scanning methods are used, then depth data can be processed, but they require good lighting and textured surfaces which limits adaptability
Solution Approach 1:
The patent implements a self-service mechanism where the volumetric gradient descent algorithm automatically adapts to different environmental conditions without requiring external assistance such as good lighting or textured surfaces. The system serves itself by using the volumetric representation to inherently handle varying lighting conditions and surface properties.
Solution Approach 2:
The patent creates a universal 3D scanning system that functions across diverse environmental conditions. The volumetric approach with fixed-point math provides multi-functionality by handling various lighting scenarios and surface textures through a single unified framework, eliminating the need for specialized processing for different conditions.
3Reliability
If traditional volumetric approaches are used for fusing depth maps, then 3D models can be created, but special auxiliary data structures are required which increase device complexity
Solution Approach 1:
The patent extracts and eliminates the need for special auxiliary data structures by using a straightforward volumetric grid system. By taking out the complex auxiliary structures from the traditional approach and replacing them with a simple fixed-point volumetric representation, the patent maintains 3D model accuracy while reducing device complexity.
Solution Approach 2:
The patent inverts the traditional approach by using a fixed-point volumetric grid system instead of floating-point with auxiliary structures. This inversion simplifies the data structure requirements while maintaining or improving 3D reconstruction reliability through the gradient descent optimization method.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for three-dimensional (3D) scanning according to an embodiment of the present disclosure may include generating a plurality of depth images while a depth sensor and an object rotates with respect to the other. The method may further include: for each depth image: estimating a rotation R and a translation T for each depth image, using data of truncated signed distance function (TSDF) volume; and fusing each depth image accumulatively into the TSDF volume based on the estimated R and T.