Object Detection Using Auxiliary Region Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current visual inspection methods for detecting regions of interest (ROI) in images, especially in manufacturing, face challenges due to variations in object appearance caused by deformation, transformation, and occlusion, making it difficult for both machine learning and template matching techniques to accurately localize and classify objects, especially when labeled image data is scarce.
Innovation Solution
A method that involves specifying auxiliary regions in a template image, matching these regions with corresponding regions in the image under detection, computing a transformation function based on the matching, and applying it to localize target regions of interest, thereby reducing the complexity of object detection and eliminating the need for extensive labeled data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If template matching or machine learning methods are used to detect ROIs, then object detection can be performed, but detection accuracy deteriorates when objects undergo deformation or transformation
Solution Approach 1:
The patent segments the object detection task into two independent parts: (1) detecting auxiliary regions that remain stable under transformation, and (2) detecting target regions using a computed transformation function. This segmentation allows the system to handle appearance variations by focusing on stable auxiliary features first, then applying the learned transformation to locate target objects regardless of their appearance changes.
Solution Approach 2:
The patent introduces auxiliary regions as intermediary elements that mediate between the template image and the image under detection. These auxiliary regions serve as stable reference points that are easier to match, and the transformation function computed from them acts as an intermediary to transfer the detection task to the target regions, bridging the gap caused by appearance variations.
2Measurement precision
If machine learning methods are used for fine-grained recognition, then classification accuracy improves, but the need for extensive labeled image data increases
Solution Approach 1:
The patent performs preliminary action by computing a transformation function from auxiliary region matching before proceeding to target region detection. This preliminary transformation computation reduces the search space and simplifies the subsequent classification task, allowing for accurate fine-grained recognition with less labeled data needed for training the classifier.
Solution Approach 2:
The patent segments the detection task into auxiliary region detection and target region detection, allowing the system to use simple geometric transformation for the first part and apply a classifier only to the localized target regions. This segmentation reduces the overall complexity and data requirements compared to training a single end-to-end machine learning model for the entire image.
3Area of stationary object
If the ROI is small while the captured image is big, then the search space increases, but detection efficiency decreases
Solution Approach 1:
The patent segments the large image search space into multiple local regions based on detected auxiliary regions. By dividing the search task into smaller local searches around each auxiliary region, the system efficiently locates small target ROIs without having to search the entire large image, thus maintaining high detection efficiency.
Solution Approach 2:
The patent uses auxiliary regions as intermediary markers that guide the search process. Instead of searching the entire large image, the system first locates auxiliary regions (which are easier to find) and then uses the computed transformation to directly locate the small target ROIs in their respective local neighborhoods, dramatically reducing the effective search space.
Data Source
AI summary
This disclosure provides a method for object detection. The method comprises receiving a user input that specifies one or more first regions and one or more second regions in a template image. The one or more second regions include one or more objects of interest. The method further comprises for each of the one or more first regions discovering a third region in an image under detection corresponding to the first region in the template image by matching the image under detection with the template image. The method further comprises computing a transformation function based on the matching from each of the one or more first regions to its corresponding third region. The method further comprises applying the computed transformation function to the one or more second regions to localize one or more fourth regions in the image under detection for the object detection.