Feature Extraction Region Segmentation for Gesture Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting feature values from images, particularly for gesture recognition in controlling devices like television receivers, face challenges in efficiently isolating effective features, leading to inaccurate posture estimation and command detection.
Innovation Solution
An information processing apparatus and method that includes an inputting section, a detection section, a noticed region setting block, a restriction region setting block, and an extraction section to isolate and extract feature values from specific regions of the image, using techniques like Ridge Regression and Non-negative matrix factorization to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature values are extracted from the entire noticed region including background, then more feature values are obtained, but background noise interferes with posture estimation accuracy
Solution Approach 1:
The noticed region is segmented into multiple blocks (e.g., 3x3 grid), and feature values are extracted only from specific blocks determined to contain the human body based on gradient analysis. This segmentation approach isolates relevant features from background noise while maintaining extraction efficiency.
Solution Approach 2:
Different regions within the noticed region are assigned different processing qualities - blocks with high gradient values (indicating human body presence) receive full feature extraction, while blocks with low gradient values (background regions) have their feature values suppressed or excluded. This local quality differentiation improves accuracy without processing the entire region uniformly.
2Measurement precision
If background feature values are suppressed using NMF, then posture estimation accuracy improves, but processing time and computational complexity increase
Solution Approach 1:
The gradient calculation and block classification are performed as preliminary actions before the main feature extraction process. By pre-identifying which blocks contain the human body based on gradient analysis, the system avoids unnecessary feature extraction from background blocks, thereby reducing overall processing time while maintaining accuracy.
Solution Approach 2:
Instead of applying computationally intensive NMF to the entire noticed region, the system applies gradient-based suppression selectively to specific blocks. This partial action approach achieves sufficient background suppression without the excessive computational cost of full-region NMF processing.
3Measurement precision
If Ridge Regression is used for nonlinear map estimation, then posture estimation accuracy improves, but computational complexity increases
Solution Approach 1:
The feature extraction region is segmented into multiple blocks, and Ridge Regression is applied separately to each block rather than to the entire noticed region. This segmentation reduces the dimensionality of the nonlinear map estimation problem, making Ridge Regression computationally feasible while maintaining accuracy through localized analysis.
Solution Approach 2:
Ridge Regression is applied with local quality consideration - full regression analysis is performed only on blocks identified as containing human body parts, while background blocks receive simplified or suppressed processing. This local application of the complex algorithm reduces overall computational complexity while preserving accuracy where needed.
Data Source
AI summary
An information processing apparatus includes: an inputting section adapted to input an image; a detection section adapted to detect a portion of an image pickup object from within the inputted image; a noticed region setting block adapted to set a noticed region from the detected portion; a restriction region setting block adapted to set a restriction region from the detected portion; and an extraction section adapted to extract feature values of the noticed region restricted by the restriction region.