Light Field Camera 3D Contour Reconstruction for Metal Additive Manufacturing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for 3D contour reconstruction in metal additive manufacturing (AM) are limited by complex calibration processes, position-dependent requirements, and the need for large training datasets and manual labeling, which restrict real-time monitoring and application of AI.
Innovation Solution
A method and system for 3D contour reconstruction of AM parts based on light field imaging, which involves calibrating a light field camera, constructing an EPI-UNet framework, capturing light field information, obtaining disparity maps, determining geometric optical path relationships, and performing disparity mapping to obtain 3D contour information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If binocular cameras are used for 3D reconstruction, then two perspectives can be provided, but complex pre-calibration process and re-calibration requirements are introduced
Solution Approach 1:
The patent uses a light field camera that captures multiple perspective images simultaneously in a single shot, creating virtual camera arrays that copy the function of multiple physical cameras. This eliminates the need for complex calibration between multiple devices while providing the same multi-perspective 3D reconstruction capability.
Solution Approach 2:
The patent merges multiple perspective views into a single light field capture, combining the functionality of multiple cameras into one device. The light field camera integrates multiple virtual camera perspectives in a single optical system, eliminating the need for separate calibration processes for each camera.
2Adaptability or versatility
If traditional stereo matching algorithms are used, then multi-view vision monitoring is achieved, but the method is limited and requires position information
Solution Approach 1:
The patent transitions from traditional 2D stereo matching to 4D light field processing by adding spatial and angular dimensions. The epipolar-plane-image (EPI) transformation converts the light field data into a format that enables disparity calculation without position dependency, adding dimensional information that simplifies the matching process.
Solution Approach 2:
The patent replaces traditional mechanical stereo matching algorithms with AI-based deep learning models. The EPI-UNet network automatically learns disparity patterns from light field data, substituting complex mechanical alignment and matching processes with intelligent pattern recognition that is insensitive to position changes.
3Measurement precision
If AI methods are used for 3D reconstruction, then new approaches are offered, but large training datasets and complex manual labeling are required
Solution Approach 1:
The patent implements self-supervised learning where the light field camera's inherent multi-perspective data provides its own training labels through geometric relationships. The system uses the light field data's internal consistency and epipolar geometry to automatically generate supervision signals, eliminating the need for external manual labeling while maintaining high reconstruction accuracy.
Solution Approach 2:
The patent introduces epipolar-plane-images (EPI) as an intermediary representation that bridges light field data and disparity maps. This intermediate format encodes depth information in a structured way that AI models can learn efficiently, reducing the complexity of training data preparation while maintaining measurement precision.
4Productivity
If high-speed cameras are used, then part surface information can be captured quickly, but angular information is lacking for reconstruction
Solution Approach 1:
The patent makes the light field camera multi-functional by enabling it to perform both high-speed surface capture and angular information acquisition simultaneously. The single-shot light field capture provides both the temporal resolution needed for speed and the angular diversity needed for 3D reconstruction, eliminating the trade-off between these two requirements.
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
This approach enables rapid and accurate 3D contour reconstruction of AM parts, simplifying the reconstruction process, and facilitating real-time monitoring and quality assurance in metal AM processes.
Implementation Method 1
a light field camera is calibrated to obtain an equivalent focal length and pixel values for the light field camera
Implementation Method 2
determining a geometric optical path relationship between disparity and depth based on the equivalent calibrated parameters of the light field camera
Data Source
AI summary
A method and system for 3D contour reconstruction of AM parts based on light field imaging, belonging to the field of image reconstruction technology is provided. The method includes constructing an EPI-UNet framework, where a preset light field dataset is used to construct a training set, learning labels from the disparity maps corresponding to the preset light field dataset are obtained, and EPI-UNet framework is trained to obtain a predicted disparity vector with the training set and learning labels. Two mappings including disparity and depth mapping, and disparity and 3D mapping, are established to get 3D contour of the AM part. The experiments of validation for accuracy and 3D contour reconstruction of AM parts were performed. By applying light field multi-view vision to the AM process and combining the rich angle and spatial domain view information of the light field, this disclosure provides a reliable quality assurance for AM monitoring.