CN-120892990-B - Old people falling risk assessment and dynamic protection intervention system and method

CN120892990BCN 120892990 BCN120892990 BCN 120892990BCN-120892990-B

Abstract

The present invention discloses an elderly fall risk assessment and dynamic protection intervention system and method, which relates to the field of computer vision technology and includes hardware composition module, software and algorithm module, intervention mechanism module, system integration and optimization module, and application scenario module. The hardware composition module is used to collect real-time motion posture, gait, and environmental data of the elderly through multimodal sensors, providing hardware support for risk assessment and protection. The present invention achieves deep fusion of visual, inertial, and environmental data through attention mechanism, combined with lightweight CNN-LSTM The network quickly extracts spatiotemporal features and accurately calculates the dynamic relationship between the center of mass and the support surface based on biomechanical models, effectively improving the real-time performance of posture analysis and the accuracy of fall risk assessment. While reducing computational resource consumption, the system greatly improves the accuracy of predicting the risk of falls in the elderly, reduces false alarm rates, and provides reliable basis for graded protection strategies.

Inventors

YANG JINXIN
WANG CEN
ZHAO HONG
FU XIAOLAN
WANG HUAN

Assignees

合肥职业技术学院

Dates

Publication Date: 20260512
Application Date: 20250723

Claims (7)

1. The old people falling risk assessment and dynamic protection intervention system is characterized by comprising a hardware composition module, a software and algorithm module, an intervention mechanism module, a system integration and optimization module and an application scene module, wherein: The hardware composition module is used for collecting the movement gesture, gait and environment data of the old through the multi-mode sensor in real time and providing hardware support for risk assessment and protection; The software and algorithm module is used for carrying out real-time analysis on the posture of the old people, prediction on the falling risk and generation of a hierarchical protection strategy by fusing multi-mode data with a lightweight intelligent algorithm, and integrating privacy protection encryption calculation; the intervention mechanism module is used for predicting the unbalance trend through biomechanics simulation and combining the MEMS microjet air bag with the reinforcement learning dynamic threshold value to carry out hierarchical accurate protection and falling emergency linkage; the system integration and optimization module is used for carrying out efficient deployment and continuous optimization through embedded edge calculation, privacy protection collaborative learning and dynamic resource management; The application scene module is used for adapting to different environments of families, nursing homes and public places and providing a customized falling protection solution; the hardware composition module comprises a visual perception unit, an inertial sensing unit, an environment perception unit, a wearable monitoring unit and a data preprocessing unit, wherein: The visual perception unit is used for acquiring three-dimensional posture coordinates of a human body and environment semantic information through the depth camera and the RGB camera; The inertial sensing unit is used for collecting motion parameters of acceleration and angular velocity through the IMU array and compensating vision shielding errors; The environment sensing unit is used for detecting environment risk factors of ground smoothness and obstacle distribution by utilizing millimeter wave radars and infrared sensors; The wearable monitoring unit is used for integrating a flexible pressure sensor to monitor plantar pressure distribution and gravity center change; The data preprocessing unit is used for carrying out noise reduction, format conversion and synchronization processing on the original sensor data and providing standardized data for subsequent analysis; The software and algorithm module comprises a multi-mode fusion unit, a lightweight analysis unit, a risk prediction unit, a strategy generation unit and a privacy calculation unit, wherein: The multi-mode fusion unit is used for weighting and fusing vision, IMU and environment data by adopting an attention mechanism to generate a motion representation with consistent space and time; The light weight analysis unit carries out real-time attitude estimation and gait cycle analysis based on the CNN-LSTM network after distillation pruning; the risk prediction unit is used for calculating the relation between the centroid projection and the supporting surface through the biomechanical model and outputting falling probability; The strategy generation unit is used for generating a corresponding hierarchical protection strategy according to the risk assessment result and guiding the intervention mechanism to execute; the privacy calculating unit is used for finishing data desensitization and encryption aggregation at the edge end by adopting a federal learning framework; The intervention mechanism module comprises a risk pre-judging unit, a protection executing unit, a dynamic tuning unit and an emergency linkage unit, wherein: the risk pre-judging unit is used for analyzing the action trend of the old in real time based on the biomechanics simulation model and pre-judging the unbalance risk; The protection execution unit is used for triggering the MEMS microjet air bag to perform physical buffering protection according to the hierarchical protection strategy; The dynamic tuning unit is used for dynamically tuning the risk threshold according to the historical data of the user by using a PPO reinforcement learning algorithm; The emergency linkage unit is used for automatically triggering GPS positioning alarm after falling confirmation and pushing medical files to an emergency center.
2. The fall risk assessment and dynamic protection intervention system for the elderly according to claim 1, wherein the lightweight analysis unit performs real-time posture estimation and gait cycle analysis based on a distilled pruned CNN-LSTM network, and specifically comprises the following steps: (1) And extracting spatial features, namely replacing standard convolution by depth separable convolution, wherein the calculation formula is as follows: Wherein, the For the decomposed depth convolution kernel, the parameter quantity is reduced to 1/8 of that of the standard convolution; (2) Performing time sequence modeling, namely performing channel pruning on an LSTM layer, and compressing the number of hidden layer units from 256 to 128; (3) Knowledge distillation, namely outputting an attitude angle thermodynamic diagram by a teacher model as a soft label to guide training of a student model, wherein a loss function is as follows: Wherein the method comprises the steps of As a weight of the task(s), Loss of KL by distillation.
3. The fall risk assessment and dynamic protection intervention system for the elderly according to claim 1, wherein the privacy calculation unit adopts a federal learning framework, and data desensitization and encryption aggregation are completed at the edge end, and the specific operation is as follows: (1) Data desensitization, namely adopting k-anonymization processing to local training data to ensure that at least 5 records are indistinguishable; (2) And (3) encryption aggregation, namely aggregating model parameters by adopting a Paillier homomorphic encryption algorithm, wherein the aggregation process meets the following conditions: Wherein, the In order to encrypt the parameters of the data, Is a public key modulus; (3) Differential privacy-injecting Laplace noise when global model updates Noise scale Privacy budgets 。
4. The fall risk assessment and dynamic protection intervention system for the elderly according to claim 1, wherein the protection execution unit triggers the MEMS microjet air bag to perform physical buffering protection according to a hierarchical protection strategy, and the specific operation is as follows: (1) The gas generation mechanism adopts a solid sodium azide micro blasting device, and generates within 20ms after triggering Nitrogen gas; (2) The multi-region control is that the inflation pressure of the 6 air bag regions at the waist and the hip is independently controlled to be 20-30kPa according to the prediction result of the falling direction; (3) Linkage braking, namely, the air bag triggering synchronously sends PWM braking signals to the walker to output reverse torque 。
5. The fall risk assessment and dynamic protection intervention system for elderly people according to claim 1, wherein the system integration and optimization module comprises an edge calculation unit, a collaborative learning unit, a resource scheduling unit and a human-computer interaction unit, wherein: The edge computing unit supports INT8 quantitative reasoning through Jetson Xavier NX deployment models; the collaborative learning unit is used for performing cross-mechanism model updating by adopting federal learning with differential privacy protection; the resource scheduling unit is used for dynamically distributing computing power and storage resources according to task priorities; The man-machine interaction unit is used for providing voice prompt, vibration feedback and AR interface guidance.
6. The system and method for senile fall risk assessment and dynamic protection intervention as set forth in claim 1, wherein the application scenario module comprises an environment adaptation unit, an organization management unit, a public deployment unit, and a personalized configuration unit, wherein: the environment adapting unit is used for adjusting the deployment scheme of the device according to different layouts and requirements of families, nursing homes and public places; The mechanism management unit is used for deploying a multi-node monitoring network, and the central nursing platform displays the group risk thermodynamic diagram and the early warning statistics in real time; the public deployment unit is used for realizing low-cost transformation by utilizing the existing monitoring camera and an edge AI box and covering important areas of a toilet and stairs; the personalized configuration unit is used for customizing the protection level through the mobile terminal APP.
7. A method for evaluating and dynamically protecting and intervening the fall risk of the elderly, which is based on the fall risk evaluation and dynamically protecting and intervening system of any one of claims 1-6, and is characterized by comprising the following steps: S1, synchronously acquiring motion gesture, gait and environment data in real time through a depth camera, an IMU array and a millimeter wave radar, and performing space-time alignment and noise reduction treatment; S2, weighting and fusing multi-mode data by adopting an attention mechanism, estimating the real-time gesture based on a lightweight CNN-LSTM network, calculating the relation between the mass center and the supporting surface through a biomechanical model, and outputting the hierarchical falling probability; s3, dynamically generating a protection strategy according to the falling probability, and triggering MEMS micro-jet air bag multi-partition pressurizing and walker linkage braking when the risk level exceeds a threshold value; s4, aggregating encryption model parameters through a federal learning framework, injecting epsilon less than or equal to 0.5 differential privacy noise, and dynamically adjusting a risk judgment threshold value by utilizing PPO reinforcement learning; s5, according to environmental characteristics of families, nursing homes or public places, a sensor deployment scheme and a protection response strategy are adaptively adjusted, and personalized configuration is achieved through the mobile terminal APP.

Description

Old people falling risk assessment and dynamic protection intervention system and method Technical Field The invention relates to the technical field of computer vision, in particular to a fall risk assessment and dynamic protection intervention system and method for old people. Background Elderly people experience a fall event, wherein about 20% of falls can cause serious injuries, such as hip fracture, craniocerebral trauma, etc., not only significantly reduce the quality of life of the elderly, but also bring about a heavy medical burden and socioeconomic pressure. Medical research shows that the fall of the old is mainly caused by physical function decline (such as muscle strength weakening and balance sense decline), chronic diseases (such as Parkinson's disease and cardiovascular and cerebrovascular diseases), drug side effects (such as dizziness caused by antihypertensive drugs), environmental factors (such as ground wet and slippery, dim light) and the like. At present, although the technology for detecting and protecting the old people from falling has a certain progress, the technology still has the obvious defects that the traditional falling detection technology has the problems of easy false alarm of environmental interference, high cloud processing delay, lack of active protection capability and high consumption of computing resources. Based on the above, the invention provides a system and a method for fall risk assessment and dynamic protection intervention of old people. Disclosure of Invention The invention aims to solve the problems in the prior art, and provides a fall risk assessment and dynamic protection intervention system and method for old people. The old person fall risk assessment and dynamic protection intervention system comprises a hardware composition module, a software and algorithm module, an intervention mechanism module, a system integration and optimization module and an application scene module, wherein: The hardware composition module is used for collecting the movement gesture, gait and environment data of the old through the multi-mode sensor in real time and providing hardware support for risk assessment and protection; The software and algorithm module is used for carrying out real-time analysis on the posture of the old people, prediction on the falling risk and generation of a hierarchical protection strategy by fusing multi-mode data with a lightweight intelligent algorithm, and integrating privacy protection encryption calculation; the intervention mechanism module is used for predicting the unbalance trend through biomechanics simulation and combining the MEMS microjet air bag with the reinforcement learning dynamic threshold value to carry out hierarchical accurate protection and falling emergency linkage; the system integration and optimization module is used for carrying out efficient deployment and continuous optimization through embedded edge calculation, privacy protection collaborative learning and dynamic resource management; the application scene module is used for adapting to different environments of families, nursing homes and public places and providing a customized falling protection solution. Preferably, the hardware composition module comprises a visual perception unit, an inertial sensing unit, an environment perception unit, a wearable monitoring unit and a data preprocessing unit, wherein: The visual perception unit is used for acquiring three-dimensional posture coordinates of a human body and environment semantic information through the depth camera and the RGB camera; The inertial sensing unit is used for collecting motion parameters of acceleration and angular velocity through the IMU array and compensating vision shielding errors; The environment sensing unit is used for detecting environment risk factors of ground smoothness and obstacle distribution by utilizing millimeter wave radars and infrared sensors; The wearable monitoring unit is used for integrating a flexible pressure sensor to monitor plantar pressure distribution and gravity center change; The data preprocessing unit is used for carrying out noise reduction, format conversion and synchronization processing on the original sensor data and providing standardized data for subsequent analysis. Preferably, the software and algorithm module comprises a multi-mode fusion unit, a lightweight analysis unit, a risk prediction unit, a policy generation unit and a privacy calculation unit, wherein: The multi-mode fusion unit is used for weighting and fusing vision, IMU and environment data by adopting an attention mechanism to generate a motion representation with consistent space and time; The light weight analysis unit carries out real-time attitude estimation and gait cycle analysis based on the CNN-LSTM network after distillation pruning; the risk prediction unit is used for calculating the relation between the centroid projection and the supporting surface through the biomechanic