Motion Feature Extraction for Accurate Pose Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current motion recognition technologies for user pose estimation in u-healthcare applications face challenges in efficiently and accurately analyzing, classifying, and evaluating user motions for indoor sports, home training, and posture corrections, as they require improved efficiency and accuracy in motion analysis using skeleton information.
Innovation Solution
A method and apparatus that extract motion feature data from video captures, calculate dynamic time warping (DTW) values to compare with model motions, and visualize similarities in radial charts, allowing for efficient and accurate classification and evaluation of user motions by selecting the most similar model motion based on these features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion analysis is performed using skeleton information from video data, then user pose estimation can be achieved without separate sensors, but the efficiency and accuracy of motion classification and evaluation need improvement
Solution Approach 1:
The motion analysis process is segmented into distinct stages: video data acquisition, skeleton information extraction, motion feature generation, and model motion comparison. This segmentation allows each stage to be optimized independently, improving both accuracy and efficiency of the overall motion classification system
Solution Approach 2:
Motion feature data serves as an intermediary between raw skeleton information and model motion comparison. By extracting meaningful features (distances between joints, bounding box dimensions) as intermediate representations, the system achieves more accurate and efficient motion classification than direct skeleton comparison
2Measurement precision
If detailed skeleton information is extracted and analyzed, then motion recognition accuracy improves, but computational complexity increases
Solution Approach 1:
The patent extracts only the essential motion features from complete skeleton information - specifically distances between key joints and bounding box dimensions. This selective extraction maintains motion recognition accuracy while significantly reducing computational complexity by processing only relevant features rather than all skeleton data
Solution Approach 2:
Instead of processing raw skeleton coordinates directly, the system creates simplified copies in the form of motion feature data (distances and dimensions). These feature copies preserve the essential motion characteristics while being computationally more efficient to compare and analyze
Data Source
AI summary
The present disclosure relates to a method and apparatus for classifying and evaluating a motion based on a motion feature. The method of classifying and evaluating a motion based on a motion feature includes obtaining video data by capturing a first motion, obtaining information on locations of joints of the first motion based on the video data, generating motion feature data of the first motion based on the information on the locations of the joints, and selecting a model motion most similar to the first motion, among a plurality of model motions, based on the motion feature data of the first motion and motion feature data of the plurality of model motions.