Image Recognition Device Margin Region Expansion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image recognition technologies struggle to recognize objects when parts of the object protrude outside the image frame, leading to inaccurate detection and increased data storage requirements for various reference images.
Innovation Solution
The solution involves expanding the detection window region by adding a margin region with predetermined image data around the captured image, allowing the recognition of objects even when they protrude beyond the original frame, and using Multi-Resolution CoHOG (MRCoHOG) technology for feature extraction and comparison.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the detection window is limited to the image frame boundaries, then the device complexity is reduced, but the recognition accuracy deteriorates when objects protrude outside the image
Solution Approach 1:
The patent extends the detection space from the two-dimensional image frame to a three-dimensional region by adding a margin region outside the image boundaries. This allows detection windows to probe beyond the visible image area, capturing objects that protrude outside the frame while maintaining systematic processing through structured padding with transparent or background-matched data.
2Measurement precision
If multiple reference images covering all possible protruding positions are stored, then the recognition accuracy is improved, but the data storage requirement increases enormously
Solution Approach 1:
The patent creates a universal reference image database that covers complete objects without protrusions, which can be applied to all detection scenarios including those where objects protrude outside the image. By combining these universal references with margin region detection and partial matching algorithms, the system handles protruding objects without needing separate reference images for each possible protrusion position, thus maintaining recognition accuracy while avoiding enormous data storage requirements.
3Measurement precision
If the detection window size is increased to capture protruding objects, then the recognition accuracy is improved, but the processing speed deteriorates
Solution Approach 1:
The patent divides the detection process into multiple stages: first scanning with smaller detection windows within the image frame, then selectively applying larger detection windows in margin regions only when protrusion is detected or suspected. This segmented approach maintains processing speed by avoiding exhaustive large-window scanning while improving recognition accuracy for protruding objects through targeted extended detection.
Data Source
Figure 1
Figure 2(a)~2(c)
Figure 3(a)~3(c)
AI summary
To detect a recognition target even if a part of the recognition target protrudes to the outside of an image. An image recognition device adds a margin region 10 filled with predetermined image data to a periphery of a captured image 1 captured by a camera to create an expanded image 11 larger than the captured image 1. When a person is too close to the camera, a part of a person image 5 protrudes from the captured image 1, but a large detection window 3e which also includes the margin region 10 is set, whereby a window image 4e including a protruding region is taken out of the expanded image 11. The window image 4e lacks a protruding part of the person image 5, but it is an image showing an entire body of the person image 5. The image recognition device stores many reference images 7 assuming various states of a person 8 (states in which positions, directions, and others of hands or feet differ with respect to each of postures, e.g., walking, running, jumping, or sitting). The image recognition device extracts features from the window image 4e, and compares them with a feature of the reference image 7, thereby recognizing the person image 5.