Target Detection via Temporal Scene Change Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing foreground object detection systems in computer vision are brittle and prone to failure due to clutter, especially shadows, and are not effective in dynamic scenes or varying conditions, leading to high false detection rates and missed detections.
Innovation Solution
A method and system that detects targets by analyzing temporal changes using unsupervised learning for intensity distribution and supervised learning for pixel identification, employing joint hue and intensity histograms, and residual normal flow fields to differentiate between true changes and clutter, with a coarse-to-fine image registration process for alignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional change detection methods are used to detect foreground objects, then target detection capability is provided, but false detection rate increases due to clutter especially shadows
Solution Approach 1:
The patent segments the change detection process into multiple stages: initial change detection to identify candidate regions, shadow detection to classify regions as shadow or target, and refined change detection only in non-shadow regions. This segmentation allows the system to handle shadows and targets separately, preventing shadows from causing false detections while maintaining sensitivity to actual targets.
Solution Approach 2:
The patent introduces shadow detection as an intermediary step between initial change detection and final target identification. This intermediary process uses color space transformation and shadow classification to filter out shadow regions before they can cause false alarms, while preserving genuine target regions for further analysis.
2Measurement precision
If background modeling is used for foreground detection, then detection is effective in static scenes, but the method fails in dynamic scenes with little overlap between images
Solution Approach 1:
The patent transitions from static background modeling to a dynamic approach that processes each image pair independently through change detection. This dynamic method does not require a pre-built background model or significant temporal overlap between images, making it adaptable to dynamic scenes and airborne imagery where the scene changes rapidly between frames.
Solution Approach 2:
The patent employs color space transformation as a parameter change strategy, converting images from RGB to a different color space that facilitates shadow detection. This parameter transformation allows the system to distinguish shadows from targets based on color characteristics rather than relying on temporal background consistency.
3Measurement precision
If color space transformation is used for shadow removal, then shadow detection capability is improved, but image noise increases in transformed space
Solution Approach 1:
The patent applies shadow detection and removal only to regions identified as shadow candidates through initial change detection, rather than processing the entire image. This partial action approach limits the propagation of noise amplification to only necessary regions, preserving image quality in non-shadow areas while still achieving effective shadow removal where needed.
Data Source
AI summary
A system and method for detecting a target in imagery is disclosed. At least one image region exhibiting changes in at least intensity is detected from among at least a pair of aligned images. A distribution of changes in at least intensity inside the at least one image region is determined using an unsupervised learning method. The distribution of changes in at least intensity is used to identify pixels experiencing changes of interest. At least one target from the identified pixels is identified using a supervised learning method. The distribution of changes in at least intensity is a joint hue and intensity histogram when the pair of images pertain to color imagery. The distribution of changes in at least intensity is an intensity histogram when the pair of images pertain to grey-level imagery.