Behavior Recognition via Dominant Direction Gradient Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current behavior recognition methods in computer vision struggle with accurately identifying behaviors from input images, especially in complex scenarios with varying lighting and noise, due to limitations in feature extraction and classification techniques.
Innovation Solution
The method involves determining gradient distributions and dominant directions in input images to detect feature points, generate feature vectors, and map these vectors to codewords in a codebook, allowing for histogram generation and classification based on learned parameters, which enhances the recognition of behaviors and objects within images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional feature extraction methods are used, then the process is simple, but the recognition accuracy deteriorates in complex scenarios with varying lighting and noise
Solution Approach 1:
The patent segments the feature extraction process into multiple stages: gradient calculation, dominant direction determination, feature point detection based on coherence contrast, and feature vector generation. This segmented approach improves recognition accuracy by capturing more discriminative information while keeping each stage computationally manageable
Solution Approach 2:
The patent changes the parameter space by using dominant direction analysis and coherence contrast instead of conventional gradient magnitude alone. By analyzing the dominant direction of gradients and computing coherence contrast based on eigenvalue differences, the method extracts more robust features that are invariant to lighting and noise variations
2Measurement precision
If detailed feature analysis is performed, then recognition accuracy improves, but processing time increases
Solution Approach 1:
The patent performs preliminary dominant direction analysis on gradient distributions before detecting feature points. By pre-computing the dominant direction and coherence contrast for each pixel, the method identifies candidate feature points more efficiently, reducing the number of pixels that require detailed analysis and thus reducing overall processing time
Solution Approach 2:
The patent applies different processing strategies to different regions: pixels with high coherence contrast are identified as feature points and undergo detailed analysis, while other pixels are processed more simply. This local quality approach ensures high detection accuracy for important features while minimizing processing time for less significant regions
Data Source
AI summary
Example embodiments disclose a method of generating a feature vector, a method of generating a histogram, a learning unit classifier, a recognition apparatus, and a detection apparatus, in which a feature point is detected from an input image based on a dominant direction analysis of a gradient distribution, and a feature vector corresponding to the detected feature point is generated.