Motion Feature Image Encoding for Human Skeleton Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The large amount of human skeleton data consumed significant storage and calculation resources in action recognition technologies, necessitating a method to reduce these resource requirements while maintaining recognition accuracy.
Innovation Solution
The method encodes human skeleton data into motion feature images, specifically forming motion feature matrices and encoding them into images, which reduces storage and calculation resources by representing data in a more compact form, such as linear or angular velocity matrices, and then processing these images using CNN models for action recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If human skeleton data is stored in traditional formats (action feature vector sequences), then recognition accuracy can be maintained, but storage resources and calculation resources are greatly consumed
Solution Approach 1:
The patent transforms human skeleton data from traditional action feature vector sequences into motion feature images by changing the data representation parameters. This involves converting joint point coordinates into motion features (such as velocity, acceleration, or other motion characteristics) and then encoding these features into image format, which fundamentally changes how the data is structured and stored while maintaining the essential information needed for recognition
Solution Approach 2:
The patent creates a compact representation (motion feature image) that copies the essential information from the original human skeleton data. Instead of storing all the detailed joint point coordinates and action feature vectors, the system creates an encoded image copy that preserves the necessary motion information while occupying significantly less storage space and requiring fewer computational resources for processing
2Measurement precision
If human skeleton data is processed in detailed formats, then recognition precision is improved, but calculation resources are greatly consumed
Solution Approach 1:
The patent extracts only the essential motion features from the complete human skeleton data. Instead of processing all joint point coordinates and detailed action features, the system extracts key motion characteristics (such as velocity, acceleration, or other motion parameters) and encodes them into motion feature images. This extraction process retains the critical information needed for accurate recognition while eliminating redundant data that consumes computational resources
Solution Approach 2:
The patent transforms the data from a traditional multi-dimensional feature vector representation into a two-dimensional image format. This dimensional change allows the application of efficient image processing techniques and neural network architectures that are optimized for image data, thereby reducing computational complexity while maintaining recognition precision
3Adaptability or versatility
If multiple groups of human skeleton data are stored separately, then action recognition can be performed, but storage resources are greatly consumed
Solution Approach 1:
The patent merges multiple groups of human skeleton data into a single motion feature image. Instead of storing each group of skeleton data separately as individual action feature vector sequences, the system encodes multiple groups into one consolidated motion feature image. This merging process preserves the information from all groups while significantly reducing the total storage space required
Data Source
AI summary
This application discloses image coding methods and apparatuses. One method comprises obtaining a plurality of groups of human skeleton data associated with performing an action by a human body, wherein each group of the plurality of groups of human skeleton data comprises joint data associated with a joint for performing the action. Based on joint data comprised in at least a portion of the plurality of groups of human skeleton data, a motion feature corresponding to the plurality of groups of human skeleton data is extracted, and the motion feature is encoded to obtain a motion feature image.