Structured Light 3D Scanning with Neural Network Scatter Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional 3D object scanners using structured light face challenges in accuracy and ease of use, particularly when scanning objects with varying surface types and surfaces that scatter light unpredictably, such as transparent or semi-transparent materials, due to limitations in image processing and sensitivity to environmental light.
Innovation Solution
A method employing a trained image processing network to improve the estimation of structured light location on the object's surface, using a combination of linear and non-linear operators to suppress light scatter effects, and generating a 3D model without requiring extensive user input for adjusting operational conditions, allowing for more accurate and efficient scanning across a wider range of objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image processing methods are used to locate structured light on object surfaces, then the scanning process is simple and fast, but the accuracy deteriorates when scanning objects with varying surface types and light-scattering materials
Solution Approach 1:
The image processing network is trained in advance on a diverse dataset of light patterns across various surface types before actual scanning. This preliminary training enables the network to automatically adapt to different surface characteristics without requiring real-time adjustment or user intervention during scanning operations.
Solution Approach 2:
The system changes the processing approach from conventional fixed algorithms to a trained neural network that can dynamically adjust its parameters based on the specific surface characteristics being scanned. The network learns optimal processing parameters for different surface types during training and applies them automatically during scanning.
2Measurement precision
If extensive user input is required to adjust operational conditions for accurate scanning, then scanning accuracy can be improved for specific object types, but the ease of operation deteriorates
Solution Approach 1:
The image processing network performs self-adjustment by automatically adapting its processing parameters based on the characteristics of the scanned object. The system serves itself by making real-time decisions about optimal processing settings without requiring user intervention or manual configuration.
Solution Approach 2:
All necessary adjustments and adaptations are performed in advance during the training phase. When actual scanning occurs, the pre-trained network immediately applies the appropriate processing strategies without requiring any user input for parameter adjustment.
3Reliability
If conventional scanning methods are used on objects with light-scattering surfaces, then the scanning process is fast and simple, but the reliability of the 3D model deteriorates
Solution Approach 1:
The system replaces conventional mechanical and algorithmic image processing approaches with a trained neural network-based processing system. This substitution enables the system to handle complex light-scattering surfaces that would be difficult or impossible to process accurately with traditional methods.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method enables accurate 3D scanning of objects with complex surfaces and reduces the time-consuming trial-and-error process, improving scanning results and making the technology more accessible to a broader range of users by providing improved accuracy and efficiency in generating point clouds or 3D models.
Implementation Method 1
a light scatter region illuminated by the structured light being projected about target positions on the surface of the physical object
Implementation Method 2
using a camera arranged at a distance from the light projector and at a viewing angle with respect to the light projector, recording a sequence of first images of at least a portion of the surface of the physical object including a light scatter region illuminated by the structured light
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A method of scanning a 3D object, comprising: via a user interface (313), receiving a user's input to perform a scanning operation and in response thereto: using a light projector (101), projecting (303) structured light onto the surface of a physical object (108) about target positions on the surface of the physical object; and using a camera (104), recording (305) a sequence of first images of at least a portion of the surface of the physical object (108) including a light scatter region illuminated by the structured light being projected about the target positions on the surface of the physical object (108); retrieving a trained image processing network (306) configured during training to output data comprising a representation of positions being estimates of the target positions on the surface of the physical object (108) in response to receiving one or more first images, including at least image intensity from the light scatter regions, in the sequence of images; using the trained image processing network (306), processing first images including the light scatter regions in the sequence of images, to generate an estimate of the target positions on the surface of the physical object (108); and using triangulation, based on the estimate of the target positions on the surface of the physical object, generating a partial or complete computer-readable, point cloud or three-dimensional model of the surface of the physical object. A deep learning convolutional neural network may be used. There is also provided a method of training such a network.