Image Recognition Method for Face-Operational Part Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition technologies face challenges in accurately associating faces and operational parts, such as hands, in images, especially when they are distant, leading to low accuracy due to the limitations of deep learning models and body key point algorithms.
Innovation Solution
An image recognition method that performs image detection to obtain face, operational part, and trunk detection results, combining these results to form multiple combinations, and then associates them to determine the relationship between faces and operational parts, improving accuracy by using feature sequences and classification models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deep learning model is used to deduce the relation between face and operational part, then the system can process complex relationships, but the accuracy decreases when the face is distant from the operational part
Solution Approach 1:
The patent segments the association task into two independent detection stages: first detecting face-keypoint pairs and operational part-keypoint pairs separately, then matching them through keypoint correspondence. This segmentation allows each detection stage to focus on local relationships, improving accuracy even when face and operational part are distant.
Solution Approach 2:
The patent introduces body keypoints as an intermediary element that mediates the association between face and operational part. By detecting keypoints on both the face region and operational part region and matching them through correspondence, the system bridges the gap between distant elements, maintaining high accuracy regardless of distance.
2Productivity
If association algorithm of body key points is used, then the processing speed is improved, but the accuracy decreases due to loss of body information
Solution Approach 1:
The patent applies partial action by focusing detection only on relevant regions (face region and operational part region) rather than the entire body. This selective detection maintains processing speed while preserving critical local information that would be lost in full-body keypoint detection, thereby improving association accuracy.
3Adaptability or versatility
If multiple faces and multiple operational parts are detected, then the system can handle complex scenes, but the difficulty of determining relationships increases
Solution Approach 1:
The patent segments the complex scene into independent face-keypoint and operational part-keypoint detection tasks. By detecting keypoints for each face and operational part separately and then matching them through correspondence, the system reduces the complexity of determining relationships in multi-person scenes while maintaining the ability to handle complex scenarios.
Data Source
AI summary
An image recognition method includes: performing image detection on an image to be recognized to obtain at least one face detection result, at least one operational part detection result, and at least one trunk detection result, each face detection result including one face bounding box, each operational part detection result including one operational part bounding box, and each trunk detection result including one trunk bounding box; respectively combining each of the at least one trunk detection result with each face detection result, to obtain at least one first result combination; respectively combining each trunk detection result with each operational part detection result, to obtain at least one second result combination; and associating the at least one first result combination with the at least one second result combination, to obtain an association result.