Global Coordinate Transformer for User Activity Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing user activity recognition techniques fail to accurately recognize user activities in real-time regardless of how the mobile terminal is being held, as they rely on device-specific coordinate systems that change with the user's holding position, causing ambiguity and limiting the effectiveness of feature extraction.
Innovation Solution
A user activity recognition apparatus and method that utilizes a Global Coordinate Transformer to normalize acceleration data from a device coordinate system to a global coordinate system, combined with a Fast Fourier Transformer to convert time-domain signals to frequency-domain signals, and machine learning models for feature extraction and classification, enabling robust activity recognition across different holding positions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If device coordinate system is used for activity recognition, then the system is simple to implement, but the recognition accuracy deteriorates when the mobile terminal is held in different positions
Solution Approach 1:
The patent introduces a coordinate transformation module as an intermediary between the acceleration sensor (in device coordinate system) and the activity recognition algorithm. This module transforms acceleration data from the device coordinate system to a global coordinate system, enabling accurate activity recognition regardless of how the mobile terminal is held, while maintaining implementation feasibility through standardized transformation procedures
Solution Approach 2:
The patent changes the reference frame parameter from device-specific coordinate system to global coordinate system. By transforming the coordinate system parameter, the system maintains recognition accuracy across different holding positions without requiring complex re-calibration or multiple sensors
2Measurement precision
If coordinate transformation and frequency domain analysis are added, then activity recognition accuracy improves, but system complexity increases
Solution Approach 1:
The patent performs coordinate transformation as a preliminary step before activity recognition, converting acceleration data to global coordinate system in advance. This preliminary action eliminates the need for complex real-time adjustments during recognition, simplifying the overall processing pipeline while maintaining high accuracy
Solution Approach 2:
The patent replaces complex mechanical or algorithmic adaptations for different holding positions with a mathematical coordinate transformation system. This substitution uses straightforward linear algebra operations instead of complex adaptive algorithms, reducing computational complexity while improving robustness
Data Source
AI summary
A user activity real-time recognition apparatus and method are provided and include a collector configured to collect a frequency-domain signal for each user activity and to generate learning data based on the frequency-domain signal. The apparatus and method also include an extractor configured to extract a user activity feature from the frequency-domain signal based on an activity feature extracting model. The activity feature extracting model is learned based on the learning data from the collector. The apparatus and method further include a classifier configured to analyze the user activity feature to classify a user activity pattern based on an activity pattern classifying model and configured to transmit the classified user activity pattern to an application device.