Compressive Sensing Machine Vision for Sparse 3D Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine vision systems require capturing a 2-dimensional intensity image of substantial size for each line of physical coordinates, making the process of forming 3D images excessively time-consuming, especially when dealing with sparse illumination, which is too slow for many industrial applications.
Innovation Solution
The system employs a method where sets of control signals are applied to rows of a pixel array, with each set being uncorrelated and representative of a different vector from a matrix product of a random basis function and a filtering function, allowing for the aggregation and digitization of output signals from columns of pixel elements to form a measurement indicative of the image, thereby reducing the number of measurements needed and improving throughput.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional 2D intensity image capture is used for each line of physical coordinates, then measurement precision is maintained, but productivity decreases significantly (100 times slower)
Solution Approach 1:
The patent extracts only the essential information needed for 3D measurement by using compressive sensing to capture a reduced set of measurements from the full 2D image data, thereby maintaining measurement precision while dramatically improving acquisition speed
Solution Approach 2:
The patent changes the measurement parameters by using random basis functions and filtering functions to transform the measurement process from capturing complete 2D intensity images to capturing a smaller set of processed measurements that still contain sufficient information for accurate 3D reconstruction
2Loss of information
If complete 2D intensity images are captured for accurate scene representation, then information completeness is maintained, but loss of time increases (acquisition time 100 times longer)
Solution Approach 1:
The patent extracts the most critical scene information through compressive sensing measurements that capture essential features while discarding redundant data, thus reducing acquisition time without significant loss of useful information for 3D reconstruction
Solution Approach 2:
The patent introduces random basis functions and filtering functions as intermediaries between the scene and the measurement process, enabling efficient information extraction that reduces acquisition time while preserving necessary scene details
3Measurement precision
If laser-line based 3D image formation is used, then measurement capability is achieved, but productivity decreases (too slow for industrial applications)
Solution Approach 1:
The patent extracts only the necessary measurement data from the laser-line illumination scene using compressive sensing, capturing essential 3D information without requiring complete 2D image acquisition, thereby achieving industrial-grade throughput
Solution Approach 2:
The patent transforms the measurement approach by applying random basis functions and filtering to the laser-line imaging process, changing from complete image capture to selective measurement extraction, enabling high-speed 3D scanning suitable for industrial applications
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 significantly reduces the time required to capture 3D images by minimizing the number of measurements needed, enhancing the speed and efficiency of the machine vision system, particularly in applications with sparse illumination.
Implementation Method 1
accumulating a first pixel signal based on incoming light energy for each of a plurality of pixel elements of a pixel array, the pixel elements each including a light sensor
Data Source
Figure 1A
Figure 1B
Figure 2A
AI summary
A machine vision system to form a one dimensional digital representation of a low information content scene, e.g., a scene that is sparsely illuminated by an illumination plane, and the one dimensional digital representation is a projection formed with respect to columns of a rectangular pixel array of the machine vision system.