3D Pattern Discrimination via Stereo Pixel Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pattern recognition technologies fail to effectively discriminate three-dimensional shapes from two-dimensional image data, leading to loss of structural information and inability to describe three-dimensional features accurately.
Innovation Solution
A pattern discriminating apparatus that calculates pixel feature values in three-dimensional image data, uses a co-occurrence matrix to analyze the frequency of pixel feature value combinations, and applies a specific mapping to discriminate objects based on learned samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If two-dimensional image data is used for pattern recognition, then the processing is simple and fast, but the three-dimensional shape information is lost
Solution Approach 1:
The patent transforms two-dimensional image data into three-dimensional representation by calculating depth information through pixel displacement between left and right stereo images. The mapping unit maps pixels from the left image to the right image and calculates depth based on displacement, thereby adding the depth dimension to the original two-dimensional data without requiring complex three-dimensional imaging equipment.
2Loss of information
If reticular pattern segmentation is used for feature extraction, then the processing is efficient, but the structure information of texture is not reflected
Solution Approach 1:
The patent extends the traditional two-dimensional reticular pattern segmentation into three-dimensional space by incorporating depth information. The setting unit sets analysis areas in three-dimensional space, and the mapping unit calculates depth for each pixel, allowing the feature extraction to capture texture structure information that varies with depth while maintaining efficient processing through systematic segmentation.
3Measurement precision
If co-occurrence matrix is calculated from mapped points, then the three-dimensional structure is captured, but the calculation complexity increases
Solution Approach 1:
The patent divides the three-dimensional space into multiple analysis areas using reticular pattern segmentation. The co-occurrence matrix calculation is performed separately for each area rather than for the entire three-dimensional space at once. This segmentation approach captures three-dimensional structure information while reducing computational complexity by processing smaller, localized regions independently.
Data Source
AI summary
A pattern discriminating apparatus includes a setting unit configured to set at least one area in a three-dimensional space in a three-dimensional image data, a feature value calculating unit configured to calculate a pixel feature value from one pixel to another of the three-dimensional image data, a matrix calculating unit configured to (1) obtain at least one point on a three-dimensional coordinate in the area which is displaced in position from a focused point on the three-dimensional coordinate in the area by a specific mapping, and (2) calculate a co-occurrence matrix which expresses the frequency of occurrence of a combination of the pixel feature value of the focused point in the area and the pixel feature values of the mapped respective points, and a discriminating unit configured to discriminate whether or not an object to be detected is imaged in the area on the basis of the combination of the specific mapping and the co-occurrence matrix and a learning sample of the object to be detected which is learned in advance.