Image Search Using Keypoint Pose Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image search technologies do not effectively consider the pose of individuals in images, limiting the accuracy of searches based on human posture and movement.
Innovation Solution
A method and apparatus that utilize reference keypoint data from a reference image to search for target images containing candidates with similar poses, using pose recognition models and a keypoint database to calculate pose similarity and identify matching images, with the option to replace candidate persons with the reference person in the target images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image search is performed based on image tags or image content, then the search can be performed with existing technologies, but the accuracy of pose-based search is insufficient
Solution Approach 1:
The patent segments the image search task into multiple independent modules: pose estimation module, keypoint extraction module, similarity calculation module, and image search module. Each module processes specific information (pose data, keypoint coordinates, image features) separately before integrating results, which improves pose recognition accuracy while maintaining manageable system complexity through modular architecture
Solution Approach 2:
The patent introduces keypoint data as an intermediary element that bridges the reference image and candidate images. By extracting keypoint coordinates and pose information from the reference image and using them as query parameters to search candidate images, the system achieves accurate pose-based search without requiring complex direct comparison mechanisms between entire images
2Measurement precision
If pose similarity calculation is performed for all candidate images, then search accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary extraction of pose information and keypoint coordinates from the reference image before the actual search process. By pre-processing the reference image to obtain normalized pose data and keypoint locations, the system can quickly compare these pre-computed features against candidate images without repeatedly performing complex pose estimation, thereby reducing overall processing time while maintaining high search accuracy
Solution Approach 2:
The patent calculates pose similarity for all candidate images to ensure comprehensive search coverage, but optimizes the similarity calculation by using normalized keypoint coordinates and pose vectors that require computationally efficient comparisons. This partial excessive action approach ensures no potential matches are missed while keeping the computational cost manageable through mathematical optimizations
3Measurement precision
If keypoint data is extracted from all candidate images, then pose matching accuracy is improved, but data processing complexity increases
Solution Approach 1:
The patent extracts keypoint data from candidate images using a universal pose estimation model that can handle various image types and conditions. The same keypoint extraction algorithm and coordinate normalization process applied to the reference image are also applied to all candidate images, creating a consistent multi-functional processing pipeline that improves pose matching accuracy while avoiding the need for image-type-specific complex processing
Data Source
AI summary
Embodiments of the present disclosure provide a method and an apparatus for searching for an image and a related storage medium. The method includes obtaining reference keypoint data of a reference person in a reference image, and searching, based on the reference keypoint data, a set of candidate images for at least one target image containing at least one candidate person that has a pose similar to the reference person.