Image Detection Region Overlap for Boundary Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods using machine learning models face issues with erroneous detection and undetected states due to predetermined input size constraints, leading to increased processing load and reduced accuracy when objects are on boundary lines or not detected at all.
Innovation Solution
An information processing apparatus that sets detection regions with overlapping ranges based on detected object sizes, using a learning model to enhance accuracy and reduce processing load by integrating detection results within and between regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the image is divided into small regions to satisfy the predetermined input size of the learning model, then the detection accuracy within each region is improved, but objects on boundary lines may be erroneously detected or undetected
Solution Approach 1:
The image is divided into multiple small regions (first detection regions) to satisfy the predetermined input size requirements of the learning model, enabling accurate detection within each region while maintaining overall image coverage
Solution Approach 2:
Second detection regions are set to include overlapping ranges that encompass the boundary lines between first detection regions. These nested regions allow the system to detect objects that may be positioned on boundary lines, preventing erroneous detection or complete miss detection while maintaining the benefits of regional division
2Reliability
If detection processing is performed again on each divided region to handle boundary objects, then detection reliability is improved, but processing load increases
Solution Approach 1:
The system performs preliminary detection on the entire image using the divided first detection regions to identify candidate objects. Then, it selectively performs additional detection processing only on second detection regions that include boundary lines, avoiding redundant processing on all regions while ensuring boundary objects are properly detected
Solution Approach 2:
The system applies different detection strategies to different regions: standard detection for internal regions and enhanced detection with overlapping ranges for boundary regions. This localized approach ensures high reliability for boundary objects while maintaining processing efficiency for the overall image
3Device complexity
If the learning model uses a predetermined input size, then the model structure is simplified, but preprocessing such as reduction or division is required which increases processing complexity
Solution Approach 1:
The image is segmented into multiple small regions that each satisfy the predetermined input size of the learning model. This allows the use of a fixed-structure model without modification while efficiently processing large images through division and subsequent integration of results
Data Source
AI summary
An information processing apparatus is configured to acquire detection target sizes at individual positions in an image, set a plurality of detection regions for the image, and detect a detection target in each detection region from the image acquired. At the time of setting the detection regions, based on the detection target sizes at the individual positions, sizes of the detection regions at the individual positions in the image and overlapping ranges between adjacent detection regions are set.