Fall risk real-time evaluation system and method based on multi-scale spatio-temporal hierarchical network

Abstract

The invention relates to the field of wearable computing, in particular to a fall risk real-time evaluation system and method based on a multi-scale spatio-temporal hierarchical network. The system comprises a data preprocessing module, a hierarchical evaluation network and a batch conversion and voting mechanism module, wherein the hierarchical evaluation network comprises a first-layer fall risk evaluation module, a multi-scale spatial-temporal feature extraction module and a second-layer fall risk evaluation module, and the second-layer fall risk evaluation module comprises an adversarial domain adaptation structure and a sample fall risk output structure. According to the invention, a secondary screening opportunity is provided for high-risk samples, and the probability that the high-risk samples are missing from detection is reduced; and meanwhile, the performance of a model in the face of unknown samples in a cross-subject scene is optimized, voting is performed by using less gait data, and the purpose of real-time evaluation is achieved.

Description

technical field [0001] The invention relates to the field of wearable computing, in particular to a real-time fall risk assessment system and method based on a multi-scale spatio-temporal hierarchical network. Background technique [0002] Falls, as one of the most common causes of accidental injuries among the elderly worldwide, have the characteristics of high frequency of occurrence, high treatment costs, and long recovery time, which seriously affect the health and daily life of the elderly. Long-term real-time monitoring of fall risk can effectively reduce the incidence of falls in the elderly and improve the quality of life of the elderly. At present, traditional methods commonly used at home and abroad to assess the risk of falls in the elderly include observation, scale questionnaires, and motor function tests. Due to the influence of venues and costs, it is impossible to evaluate the risk of falls in the elderly in real time for a long time. Sensors and wearable te...

AI Technical Summary

Problems solved by technology

[0003] Deep learning has made a breakthrough in the real-time performance of low-gait assessment of fall risk. The current machine learning method needs to extract artificial features from multi-gait data, which is insufficient in real-time performance and requires prior knowledge and experience

Existing fall risk assessments based on deep learning methods mostly use inertial sensors (IMU) deployed in multiple parts of the body, which is not conducive to long-term wear, low comfort, and does not pay attention to the hazards of missed detection of high-risk samples; some methods do not Considering the application scenarios of cross-subjects, it is impossible to ensure the same excellent performance in the face of unknown samples; at the same time, there is currently no way to model the fall risk assessment model of plantar pressure in cross-subject scenarios, and the advantages of foot pressure monitoring are not fully utilized. Association of comfort, portability, and foot pressure information with fall risk

Images