Resident Activity Recognition via Sensor Weight Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current home care systems face challenges in accurately recognizing activities of multiple residents simultaneously due to increased sensor data complexity, leading to lower recognition accuracy.
Innovation Solution
A resident activity recognition system that uses a processor and memory device to differentiate between target and non-target residents by generating and updating sensor weight sets based on models trained with past data, allowing for simultaneous activity recognition without requiring residents to wear sensors or provide positioning information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If activity recognition is performed for multiple residents simultaneously, then the system can monitor all residents, but the sensor data complexity increases leading to lower recognition accuracy
Solution Approach 1:
The patent divides the activity recognition process into two distinct segments: first recognizing the non-target resident's activity, then using that information to improve target resident recognition. This segmentation allows the system to handle multiple residents by processing them in sequence rather than simultaneously, reducing data complexity at each stage
Solution Approach 2:
The patent extracts and removes the non-target resident's activity pattern from the sensor data before performing target resident recognition. By taking out the interfering non-target resident's activity information and creating updated testing data with reduced weights, the system eliminates the source of data complexity and improves recognition accuracy
2Adaptability or versatility
If sensor data from all residents is processed together, then comprehensive monitoring is achieved, but data complexity increases reducing recognition accuracy
Solution Approach 1:
The system extracts the non-target resident's activity contribution from the combined sensor data and removes it by reducing the weight of sensors associated with that resident's activity. This extraction process separates the target resident's activity signal from the mixed sensor data, reducing overall data complexity while maintaining comprehensive monitoring capability
Solution Approach 2:
The patent applies different processing qualities to different parts of the sensor data: sensors related to non-target resident activity receive reduced weights, while other sensors maintain their original weights. This local differentiation allows the system to handle complex multi-resident data while preserving recognition accuracy for the target resident
Data Source
AI summary
A resident activity recognition method is provided. The method comprises: receiving a plurality of first testing data from a plurality of sensors by a processor, wherein the first testing data includes a present weight set of sensors and present trigger statuses of sensors; determining an activity of a non-target resident at a present time by the processor according to a non-target resident model and the first testing data; reducing a part of the present weight set to generate an updated weight set and a second testing data including the updated weight set and the present trigger statuses by the processor according to the activity of the non-target resident at the present time, the first testing data and the non-target resident model; determining an activity of a target resident at the present time by the processor according to a target resident model and the second testing data.