Fall Detector Log Likelihood Ratio Algorithm
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fall detectors face a trade-off between fall detection probability and false alarm rate, failing to meet the requirement of maintaining a high detection probability above 95% while keeping false alarms below one every two months, due to their moderate reliability as expressed by the Receiver Operating Curve (ROC).
Innovation Solution
A method and device that process measurements from sensors like accelerometers and pressure sensors to extract features associated with falls, calculate log likelihood ratios, and determine if a fall has occurred using a weighted sum of these ratios, optimizing the Receiver Operating Characteristic (ROC) to improve detection accuracy and reduce false alarms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional threshold-based feature comparison is used, then the device complexity is reduced, but the fall detection probability and false alarm rate performance deteriorates
Solution Approach 1:
The patent transforms the detection approach by changing from simple threshold-based binary comparisons to log likelihood ratio calculations. This parameter transformation enables continuous, probabilistic decision-making that captures the nuances of fall detection patterns, significantly improving reliability while maintaining manageable computational complexity through structured feature extraction and weighted scoring.
2Reliability
If multiple sensor features are processed with complex reasoning, then the fall detection probability improves, but the computational complexity and processing time increases
Solution Approach 1:
The patent segments the complex detection task into distinct phases: feature extraction from sensor data, log likelihood ratio calculation for each feature, and weighted sum aggregation. This segmentation allows each component to be optimized independently and enables parallel processing of multiple features, maintaining high detection probability while managing computational complexity through modular architecture.
Solution Approach 2:
The patent introduces log likelihood ratios as an intermediary representation that bridges raw sensor features and final fall detection decisions. This intermediary transformation consolidates complex feature relationships into a unified scoring mechanism, simplifying the reasoning process while preserving the nuanced information needed for accurate fall detection.
Data Source
AI summary
A method for detecting a fall by a user includes a method of detecting a fall by a user. The method includes processing measurements obtained from one or more sensors to extract a respective value for a plurality of features associated with a fall. A respective log likelihood ratio for each of the values is determined. Whether the user has fallen is determined based on the determined log likelihood ratios.