Motion Recognition via Atomic Pose Hashing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current human body motion recognition systems based on computer vision face high computing complexity, particularly with dynamic time warping (DTW) processing, making them unsuitable for embedded platforms with limited resources.
Innovation Solution
The method involves obtaining video frames, analyzing them to generate atomic pose feature values, and using a hash function to produce hash values for motion recognition, comparing these hash values against a hash table to identify specific motions, thereby reducing computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dynamic time warping (DTW) processing is used for motion recognition, then motion recognition accuracy is improved, but computing complexity increases significantly
Solution Approach 1:
The patent segments the continuous motion recognition process into discrete atomic pose comparisons. Instead of applying complex DTW to entire motion sequences, the system breaks down motions into individual atomic poses (specific joint configurations) and compares these segmented units directly, significantly reducing computational complexity while maintaining recognition accuracy.
Solution Approach 2:
The patent extracts only the essential atomic pose features from video frames rather than processing entire image sequences. By taking out and focusing on key atomic pose elements (joint positions and configurations), the system eliminates unnecessary computational overhead while preserving the critical information needed for accurate motion recognition.
2Adaptability or versatility
If machine learning models are used for motion recognition, then recognition capability is improved, but establishment time and computing resource consumption increase
Solution Approach 1:
The patent uses a hash table as a simplified copy of motion data structures, replacing complex machine learning models. By copying and storing atomic pose hash values in a lookup table, the system achieves fast recognition without the time-consuming model training and computation required by machine learning approaches.
Solution Approach 2:
The patent changes the recognition approach from learning-based parameter optimization to direct parameter comparison. Instead of using machine learning models that require training on large datasets, the system transforms motion data into hash values and compares them directly, dramatically reducing establishment time and computing resource consumption.
3Adaptability or versatility
If template comparison with massive image templates is used, then motion recognition coverage is improved, but storage space requirements increase
Solution Approach 1:
The patent extracts only the essential atomic pose hash values from comprehensive image templates. By taking out and storing only the critical hash representation of atomic poses rather than entire image templates, the system maintains broad motion recognition coverage while dramatically reducing storage space requirements.
Solution Approach 2:
The patent changes the storage representation from massive image templates to compact hash value parameters. By transforming visual data into condensed hash representations of atomic poses, the system achieves efficient storage while preserving the ability to recognize diverse motions through parameter comparison.
Data Source
AI summary
A method, a system, and a computer-accessible recording medium for motion recognition based on an atomic pose are provided. A video frame including a live body is obtained. An atomic pose feature value is generated by analyzing the live body in the video frame. A hash value of the atomic pose is generated by executing a hash function according to the atomic pose feature value. The live body executing a specific motion is recognized by comparing the atomic pose hash value.