Inertial Sensor Activity Recognition via Dynamic Feature Space
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classifiers for determining movement activities in mobile electronic devices require large amounts of training data, leading to computationally expensive and power-intensive processes, and often result in inaccurate class definitions for individual users due to generalized classes.
Innovation Solution
A system that uses inertial sensors and processing circuitry to dynamically generate feature arrays, update existing classes, create new classes, and renormalize the state space based on new data, allowing for efficient and automatic classification of movement activities without the need for extensive training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large amounts of training data are used to train classifiers, then classification accuracy is improved, but computational cost and power consumption increase
Solution Approach 1:
The system performs classification operations in advance by maintaining pre-defined activity classes with their characteristic feature ranges. Instead of training a complex classifier on large datasets in real-time, the system pre-processes training data offline to establish class definitions, then uses these pre-computed classes for efficient real-time classification with minimal computational resources.
Solution Approach 2:
The patent extracts only the essential characteristics needed for classification by defining activity classes based on key feature ranges (e.g., acceleration thresholds, frequency bands). Rather than using the entire training dataset during runtime, the system extracts and stores only the critical class boundary information and characteristic ranges, significantly reducing the data volume required for classification while maintaining accuracy.
2Measurement precision
If large amounts of training data are used to train classifiers, then classification accuracy is improved, but device complexity and processing requirements increase
Solution Approach 1:
The patent segments the continuous activity space into discrete, non-overlapping classes with well-defined boundaries. Each activity class is divided into specific feature ranges (e.g., low, medium, high acceleration bands). This segmentation transforms a complex continuous classification problem into simpler discrete category matching, reducing algorithmic complexity while preserving classification accuracy.
Solution Approach 2:
The system changes the classification approach from complex probabilistic models to simple parameter-based threshold matching. By defining activities through explicit parameter ranges (acceleration thresholds, frequency bands, duration limits), the patent transforms the classification task into straightforward parameter comparison operations, significantly reducing computational complexity.
3Productivity
If generalized classes are used for all users, then training efficiency is improved, but individual user accuracy deteriorates
Solution Approach 1:
The patent implements dynamic class adaptation where activity class definitions can be adjusted based on individual user characteristics. The system allows class parameters (thresholds, ranges) to be modified during operation based on user-specific training data, enabling the classifier to adapt from generalized initial classes to personalized accurate classes over time.
Solution Approach 2:
The system incorporates feedback mechanisms where classification results and user corrections are used to refine individual user profiles and adjust class parameters. By continuously learning from user feedback and recalibrating class boundaries based on individual patterns, the system improves individual user accuracy while maintaining the efficiency benefits of the structured class-based approach.
Data Source
AI summary
Technological advancements are disclosed that utilize inertial sensor data associated with a device to determine a new feature array and if the new feature array is within an existing class within a state space associated with the inertial sensor data. In response to the new feature array being included in the existing class, the new feature array is added to the existing class and a representation of the existing class in the state space is updated based on the new feature array and an existing representation of the existing class. In response to the new feature array not being included in the existing class, a new class is created based on the new feature array.