Multi-Scale Image Invariant Analysis for Occlusion Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis methods for detecting objects in scenes struggle with accurately identifying transient objects, especially when they remain stationary or under varying illumination conditions, leading to corruption of the background model and failure to detect occlusions effectively.
Innovation Solution
The method employs multi-scale invariants, using contrast functions between pixel locations to detect occlusions in user-defined regions of interest, independent of temporal information, by learning stable contrast values from a reference background image, allowing for reliable detection of objects of varying sizes across different illumination conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If background model maintenance methods are used to detect objects, then detection can be performed in real-time, but the background model becomes corrupted under varying illumination conditions and when objects remain stationary
Solution Approach 1:
The patent segments the image into multiple scales using Gaussian pyramids, analyzing the image at different resolutions (octaves and scales). This multi-scale segmentation allows the system to detect objects regardless of their size and illumination conditions, as features become visible at appropriate scales in the pyramid structure
Solution Approach 2:
The patent performs preliminary action by pre-processing the image through Gaussian blurring and pyramid construction before detection. This preliminary multi-scale representation is built once and then used for detection, avoiding the need to maintain and update complex background models during operation
2Speed
If methods relying on frame-to-frame differences are used, then moving objects can be detected, but stationary objects cannot be detected
Solution Approach 1:
The patent transitions from temporal analysis (frame-to-frame differences) to spatial-frequency analysis by constructing Gaussian pyramids. This dimensional change allows detection based on spatial patterns and scale variations rather than temporal changes, enabling detection of stationary objects while maintaining sensitivity to moving objects
3Measurement precision
If pixel-based background modeling is used, then detailed object detection is possible, but illumination changes corrupt the model
Solution Approach 1:
The patent changes parameters by transforming the image into the frequency domain through Gaussian pyramids, where parameters like scale and octave level replace direct pixel intensity values. This transformation makes the detection invariant to illumination changes while preserving the ability to detect objects at different sizes and locations
Data Source
AI summary
The present invention is a method and system for detecting scene events in an imaged sequence by analysis of occlusion of user-defined regions of interest in the image. The present invention is based on the multi-scale groups of nearby pixel locations employing contrast functions, a feature that is invariant to changing illumination conditions. The feature allows the classification of each pixel location in the region of interest as occluded or not. Scene events, based on the occlusion of the regions of interest, are defined and subsequently detected in an image sequence. Example applications of this invention are automated surveillance of persons for security, and automated person counting, tracking and aisle-touch detection for market research.