Object Detection via Omnipresent Geometric Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional weakly supervised object detection technologies fail to accurately detect target object regions when they are not remarkable and when images contain regions with common appearances that are not part of the target object, leading to incorrect detection.
Innovation Solution
An object detection device that extracts local features from images, identifies common patterns across image pairs using geometrically similar feature point pairs, and detects regions based on omnipresent patterns without assuming position, size, or contrast, thereby accurately identifying target object regions even if they are not remarkable.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If conventional weakly supervised object detection technologies estimate highly-remarkable regions and associate them across image pairs, then detection can be performed without manual labeling, but detection accuracy deteriorates when target objects are not remarkable or when non-target regions have common appearances
Solution Approach 1:
The patent segments the detection process into distinct modules: a local feature extraction unit that extracts features from individual images, an image-pair common pattern extraction unit that identifies geometrically similar feature point pairs across image pairs, and a region detection unit that detects target regions based on omnipresent patterns. This segmentation allows each module to specialize in specific tasks, improving overall detection accuracy while maintaining automation.
Solution Approach 2:
Instead of detecting remarkable regions and verifying if they are target objects, the patent inverts the approach by detecting common patterns across multiple images and verifying if these omnipresent patterns correspond to target objects. This inversion fundamentally changes the detection logic from top-down (remarkable region → target verification) to bottom-up (common pattern → target confirmation), resolving the accuracy issue.
2Reliability
If close association is calculated for many image pairs to specify target object regions, then detection completeness improves, but processing cost increases
Solution Approach 1:
The patent applies partial action by selecting representative image pairs rather than calculating associations for all possible image pairs. The region detection unit identifies target regions based on common patterns in selected image pairs, which is sufficient for accurate detection without the computational burden of exhaustive pairwise comparison across the entire image set.
Solution Approach 2:
The image-pair common pattern extraction unit serves multiple functions: it extracts geometrically similar feature point pairs, identifies omnipresent patterns across images, and provides the basis for region detection. This multi-functionality reduces the need for separate processing steps and improves processing efficiency while maintaining detection completeness.
3Device complexity
If target object regions are assumed to be highly remarkable, then detection process is simplified, but detection accuracy deteriorates when objects do not meet remarkability conditions
Solution Approach 1:
The patent changes the detection parameter from remarkability (contrast, saturation, position) to geometric similarity of feature point pairs across image pairs. The image-pair common pattern extraction unit identifies patterns based on geometric relationships rather than visual prominence, allowing detection of non-remarkable objects while maintaining process simplicity through consistent feature matching.
Data Source
AI summary
Even if an object to be detected is not remarkable in images, and the input includes images including regions that are not the object to be detected and have a common appearance on the images, a region indicating the object to be detected is accurately detected. A local feature extraction unit 20 extracts a local feature of a feature point from each image included in an input image set. An image-pair common pattern extraction unit 30 extracts, from each image pair selected from images included in the image set, a common pattern constituted by a set of feature point pairs that have similar local features extracted by the local feature extraction unit 20 in images constituting the image pair, the set of feature point pairs being geometrically similar to each other. A region detection unit 50 detects, as a region indicating an object to be detected in each image included in the image set, a region that is based on a common pattern that is omnipresent in the image set, of common patterns extracted by the image-pair common pattern extraction unit 30.