Keypoint Detection Using Local Regional Proposal Boxes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current keypoint detection technologies are inadequate for complex scenes where objects overlap and are not accurate due to their reliance on ideal application scenarios with single, manually extracted objects, leading to unsatisfactory detection results.
Innovation Solution
A method and apparatus for detecting object keypoints using local regional proposal boxes, which involve extracting feature maps and employing a deep neural network to predict keypoint heatmaps and probabilities, allowing for universal object keypoint detection that is robust to background influence and inaccurate box positioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current keypoint detection technologies are used, then detection can be performed on simple scenes, but detection accuracy deteriorates in complex scenes where objects overlap
Solution Approach 1:
The patent segments the detection process into two stages: first extracting local regional proposal boxes that isolate potential object regions, then performing keypoint detection on each proposal separately. This segmentation allows the system to handle complex scenes with overlapping objects by processing each proposed region independently, thereby maintaining detection accuracy in challenging environments.
Solution Approach 2:
The patent introduces local regional proposal boxes as an intermediary step between the input image and keypoint detection. These proposal boxes act as a mediator that filters and pre-processes the image data, extracting candidate regions before keypoint detection is performed. This intermediary structure enables the system to robustly handle background interference and overlapping objects.
2Measurement precision
If manually extracted single objects are used, then detection accuracy is high for ideal scenarios, but the system becomes inadequate for complex application scenes
Solution Approach 1:
The patent creates a universal keypoint detection system that can handle multiple scene types through the use of local regional proposal boxes. Instead of requiring manually extracted single objects for each specific scenario, the system uses a generalizable proposal extraction mechanism that adapts to various complex scenes including overlapping objects, making the detection system both accurate and universally applicable.
3Measurement precision
If background influence is not inhibited, then processing is simpler, but detection accuracy deteriorates in complex application scenes
Solution Approach 1:
The patent extracts local regional proposal boxes from the full image, effectively separating the objects of interest from the background. This extraction process removes background influence by focusing computation only on relevant regions, thereby improving detection accuracy while managing processing complexity through selective region analysis rather than full-image processing.
Data Source
AI summary
A method and an apparatus for detecting an object keypoint, an electronic device, a computer readable storage medium, and a computer program include: obtaining a respective feature map of at least one local regional proposal box of an image to be detected, the at least one local regional proposal box corresponding to at least one target object; and separately performing target object keypoint detection on a corresponding local regional proposal box of the image to be detected according to the feature map of the at least one local regional proposal box.