Unsupervised fall monitoring system and method

Through the unsupervised fall monitoring system, the action signals are collected using millimeter wave radar, characteristics are extracted and time division prediction is carried out, which solves the problem of data acquisition and labeling in the existing technology, and achieves efficient and accurate fall monitoring.

CN115500809BActive Publication Date: 2025-03-18AEROSPACE INFORMATION RES INST CAS
2 Cites 0 Cited by

Patent Information

Application Number
CN202211314460.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-10-25
Publication Date
2025-03-18
Estimated Expiration
2042-10-25

AI Technical Summary

Technical Problem

Existing supervised radar fall monitoring systems have difficulties in data acquisition and labeling, especially the need to collect and mark large numbers of fall and non-fall samples.

Method used

The unsupervised fall monitoring system is adopted, and the action signals of the test objects and the monitoring objects are collected through the acquisition module. The feature extraction module extracts signal characteristics. The prediction module divides the characteristics in the time dimension and makes predictions, and calculates the total anomaly score to judge the fall.

Benefits of technology

There is no need to collect fall action data and label data, which improves monitoring efficiency, reduces errors in deep learning, and improves the accuracy of fall monitoring.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The present disclosure provides an unsupervised fall monitoring system and method thereof, which relates to the field of intelligent monitoring, and includes an acquisition module for acquiring a first motion signal of a test object and a second motion signal of a monitored object, wherein the first motion signal is a motion signal corresponding to a non-falling motion; a feature extraction module for extracting features of the first motion signal and the second motion signal; a prediction module for taking the features of the front and rear segments of the first motion signal as input, calculating the total abnormality score of the features of the middle segment of the test object, and calculating a threshold value according to the total abnormality score of the test object; taking the features of the front and rear segments of the second motion signal as input, calculating the total abnormality score of the features of the middle segment of the monitored object, and when the total abnormality score of the monitored object is greater than the threshold, the monitored object is judged to have fallen. The present disclosure has the effect of accurately detecting falls without collecting fall data and assigning labels, and only using unlabeled non-fall samples for training.
Need to check novelty before this filing date? Find Prior Art