Recursive Inertial Feature Extraction for Embedded AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting sensor features in embedded artificial intelligence mechanisms are computationally intensive and prone to missing events due to non-overlapping window analysis, leading to increased power consumption and memory usage.
Innovation Solution
A method and system that utilize recursive features calculated from sequential inertial sensor data, including recursive maximum, minimum, peak-to-peak, average, and variance, which are computed using overlapping windows to enhance event detection while reducing computational load and memory footprint.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional feature extraction methods are used, then comprehensive feature analysis is achieved, but computational complexity and power consumption increase
Solution Approach 1:
The patent extracts only the most essential features from sensor data that are sufficient for activity recognition, rather than computing all possible features. This selective extraction reduces computational load and power consumption while maintaining recognition accuracy.
Solution Approach 2:
The system computes features at a reduced rate or partial frequency compared to traditional methods, using only when necessary or at lower resolution, thereby reducing overall computational effort and energy consumption while still achieving effective activity recognition.
2Use of energy by moving object
If non-overlapping windows are used for feature detection, then computational load is reduced, but event detection accuracy deteriorates
Solution Approach 1:
The patent combines multiple non-overlapping window results through recursive feature computation, merging information from sequential windows to detect events that span across window boundaries. This approach maintains low computational load while improving event detection accuracy.
Solution Approach 2:
The system performs preliminary feature extraction on non-overlapping windows, then applies recursive calculations to combine results and detect events that occur between windows, ensuring no events are missed while maintaining computational efficiency.
3Measurement precision
If overlapping windows are used for feature detection, then event detection accuracy is improved, but memory usage and computational complexity increase
Solution Approach 1:
The patent extracts only the essential recursive features from overlapping windows rather than storing and processing all raw data points. This selective extraction reduces memory footprint while maintaining the ability to detect events accurately.
Solution Approach 2:
The system uses a reduced overlap degree between windows compared to traditional approaches, computing features only at necessary intervals. This partial overlap maintains event detection accuracy while significantly reducing memory usage and computational complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Method and system for utilizing sequential input inertial sensor data to calculate recursive features for training a machine learning algorithm or for classifying the data as a known class. The recursive feature values of a current data sample are calculating based on comparisons between the current data sample value and previous recursive feature values. The recursive features include a recursive maximum, recursive minimum, recursive peak to peak, recursive average, recursive root mean square, and recursive variance.