Marker Detection Using Unequal Side Division and Adaptive Thresholding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image recognition methods for detecting markers in images suffer from low accuracy and high processing load, particularly in varying lighting conditions and partial occlusions, and fail to accurately determine the position of markers.
Innovation Solution
An image recognition program that functions as image acquisition, vertex detection, division point generation, sample point determination, and distinction means, dividing sides of detected vertices into unequal parts based on pixel values to accurately identify markers, and includes mechanisms for correcting marker position deviations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional image recognition methods use fixed threshold binarization and pattern matching, then the processing load is reduced, but the detection accuracy deteriorates in varying lighting conditions and partial occlusions
Solution Approach 1:
The patent changes the parameter of threshold selection from fixed to adaptive based on image characteristics. Instead of using a predetermined fixed threshold, the system determines the threshold dynamically according to the actual image data, allowing accurate detection across varying lighting conditions while maintaining efficient processing
Solution Approach 2:
The patent performs preliminary analysis of image characteristics (such as histogram analysis or edge detection) before performing the main thresholding operation. This preliminary action allows the system to adapt the threshold to current conditions, improving detection accuracy without significantly increasing overall processing load
2Productivity
If conventional methods use simple connected region extraction, then the processing load is reduced, but the ability to detect markers in partial occlusion and various states deteriorates
Solution Approach 1:
The patent segments the image processing into multiple stages: connected region extraction, candidate selection based on geometric constraints, and final verification. This segmentation allows the system to handle partial occlusions and various marker states by progressively filtering candidates, improving reliability while keeping each stage computationally efficient
Solution Approach 2:
The patent introduces dynamic adaptation in the detection process by adjusting search parameters and thresholds based on detected image characteristics. The system dynamically modifies its behavior according to the actual marker states (occlusion, rotation, lighting), enhancing detection reliability without requiring exhaustive processing of all possible cases
3Adaptability or versatility
If conventional pattern matching is used on all connected regions, then detection coverage is improved, but the processing load increases significantly
Solution Approach 1:
The patent applies partial action by performing full pattern matching only on selected candidate regions that meet preliminary criteria (size, shape, position). Most connected regions are filtered out before the computationally intensive pattern matching stage, achieving adequate detection coverage with significantly reduced processing load
Solution Approach 2:
The patent applies different processing quality to different regions: simple geometric filtering to most regions and full pattern matching only to promising candidates. This local quality approach ensures adequate detection coverage where needed while minimizing processing load in other areas
4Device complexity
If conventional methods detect marker positions directly, then the processing is simplified, but position deviation occurs
Solution Approach 1:
The patent performs preliminary refinement of detected marker positions by analyzing the distribution of matched features and adjusting the position estimate accordingly. This preliminary refinement action corrects small deviations without requiring complex post-processing, maintaining simplicity while improving position accuracy
Data Source
Figure 1
Figure 2A~2B
Figure 2C~2D
AI summary
First, a plurality of vertices of a contour of an object or of a design are detected from an image. Then, a predetermined number of division points are generated on each of sides connecting the plurality of detected vertices, so as to divide each side of at least one pair of two opposing sides into unequal parts. Then, a plurality of sample points are determined on the basis of straight lines connecting the division points on the two opposing sides to one another, and on the basis of pixel values of the sample points, it is determined whether or not a predetermined object or design is displayed in an area surrounded by the plurality of vertices in the image.