CN-122024403-A - Fall-preventing monitoring and early-warning method and system based on remote monitoring

CN122024403ACN 122024403 ACN122024403 ACN 122024403ACN-122024403-A

Abstract

The application relates to the field of fall-prevention monitoring, discloses a fall-prevention monitoring early warning method and system based on remote monitoring, and aims to solve the problems of poor adaptability, high false alarm rate, delayed response and the like of the traditional fall-prevention monitoring scene. The method comprises the steps of collecting multi-source data of a target object in real time, conducting remote preprocessing and feature extraction on the multi-source data, conducting dimension reduction on motion data, completing positioning data calibration synchronously through clustering point cloud data dimensions, executing risk judgment based on a fusion inference model of a two-stage decision unit, triggering hierarchical remote early warning based on an inference result, enabling a high-risk scene to directly start acousto-optic early warning and pushing detailed information, and enabling a suspected risk scene to start a dynamic observation mode. The application realizes high accuracy, low false alarm rate and multi-scene suitability of fall monitoring, meets the requirements of non-contact privacy protection and real-time response, and can be widely applied to remote fall prevention monitoring scenes such as living families of solitary old people, monitoring in hospitals, construction sites and the like.

Inventors

SONG TING
CHEN YINHUA
Jiang Yilian
WANG WEILING
YU BAIYUAN

Assignees

安徽医科大学第一附属医院

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. The fall-preventing monitoring and early-warning method based on remote monitoring is characterized by comprising the following steps of: Acquiring multi-source data of a target object in real time, wherein the multi-source data comprises motion characteristic data, environment video data and positioning data; performing remote preprocessing and feature extraction on the acquired multi-source data, and generating an angular velocity root mean square sequence and an acceleration root mean square sequence based on the motion feature data; The system comprises a remote fusion reasoning model, a second decision unit, a first decision unit and a first analysis unit, wherein the fusion reasoning model is used for executing anti-falling monitoring early warning through the remote fusion reasoning model and comprises a two-stage decision unit, the first decision unit is used for calculating the slope angle of the angular speed root mean square sequence and triggering abnormal verification of an acceleration signal; If the local equipment and the cloud end judge that the falling risk or the falling occurs, starting acousto-optic early warning by the local equipment, and simultaneously pushing early warning information comprising falling time, position and risk level to a remote monitoring platform; If only a single level is judged to be a risk, starting a dynamic observation mode, encrypting data acquisition frequency and continuously monitoring characteristic change, wherein the single level is only a local or cloud fall risk or a fall occurs.
2. The fall prevention monitoring and early warning method based on remote monitoring according to claim 1, wherein the slope angle of the first decision unit is derived according to the following manner: ( ) , Wherein, the For the local minimum point of the angular velocity root mean square sequence in the dynamic sliding window (Min, ), Is a local maximum point (Max, ), 、 Respectively corresponding time stamps of the minimum value point and the maximum value point, Is of a slope angle when Greater than a preset threshold And triggering the abnormal verification of the acceleration signal of the second decision unit.
3. The fall prevention monitoring and early warning method based on remote monitoring according to claim 1, wherein the second decision unit performs verification by: , The similarity is calculated by the following steps: , Wherein, the Is a hash-map function for a sliding window of acceleration RMS sequences, Is the j-th preset spherically symmetric random vector, k is the total number of preset spherically symmetric random vectors, Is sliding window data of an acceleration root mean square sequence, = W is the window length, i is the serial number identification of the sliding window, wherein 3D data collected by the accelerometer firstly calculates RMS to obtain a single-dimensional RMS time sequence [ ] ); For the same number of bits of both hash values, Accumulating bits for all bits of both hash values by the same number of bits, n being the total length of the acceleration RMS time series, n=sampling frequency Sampling duration, hash value For mapping functions by k hashes (J=1, 2,., k) the bits respectively output are spliced in order; The Hamming similarity of the hash values of the adjacent windows is calculated, if the similarity is lower, the acceleration RMS sequence of the two windows is more severely changed and corresponds to the sudden change of acceleration in falling, and if the similarity is higher, the acceleration RMS sequence of the two windows corresponds to the gradual change of daily activities.
4. A method of fall prevention monitoring and early warning based on remote monitoring as claimed in claim 3, wherein the RMS calculation of the motion feature data is used for extracting the fall related motion mutation feature in a targeted manner as follows: , , Wherein, the 、、 The angular velocity values of the 3D gyroscope along X, Y, Z axes at the time t are respectively, The root mean square value of the angular velocity value of the gyroscope at the moment t; 、、 The acceleration values of the 3D accelerometer along X, Y, Z axes at the time t are respectively, The root mean square value of the acceleration at time t.
5. The fall-prevention monitoring and early-warning method based on remote monitoring as claimed in claim 1, wherein the environmental video data is clustered by: , Wherein, the As the data points in the video data, And m is the preset cluster quantity, C is a cluster set, wherein the CNN model comprises two layers of convolution layers, two layers of pooling layers and a full-connection layer, and the convolution layers respectively adopt 16 channels and 64 channels to extract features.
6. The fall prevention monitoring and early warning method based on remote monitoring according to claim 1, wherein the fusion inference model further comprises a position change analysis of positioning data and a z-axis head height speed change rate analysis of radar point cloud data, and the speed change rate is obtained according to the following manner: , Wherein, the 、 The heights of the heads of the z-axis at the time t and the time t-1 are respectively extracted based on the space coordinates of the radar point cloud data, the heights of the heads of the corresponding target objects, V is the rate of change of the z-axis head height for the time difference, when When the sudden drop, i.e. v is greater than or equal to the first set threshold and the positioning data display position does not actively move And gradually decreasing, namely, when v is in the second setting range, judging that the patient is prone.
7. The fall-preventing monitoring and early-warning method based on remote monitoring as claimed in claim 2, wherein the encrypted data acquisition frequency of the dynamic observation mode is 2-3 times of the original frequency, the duration of the monitoring is 3-5 minutes, and if the slope angle during the monitoring is the same as the original slope angle Continuously less than or equal to a preset threshold The Hamming similarity is continuously greater than the set threshold And if any feature in the monitoring period meets the risk judging condition, the system is upgraded into grading early warning.
8. The fall prevention monitoring and early warning method based on remote monitoring as claimed in claim 1, wherein the fused inference model further comprises a dynamic window adaptive adjustment, the size of the dynamic window being adjusted by a local minimum point in an angular velocity root mean square sequence And local maximum point Is set in the time interval of (2) Determining that its adjusting logic is that when The window length is enlarged by 1.5 times when the time is less than 0.2s, and the time is less than or equal to 0.2s When the time is less than or equal to 0.5s, the window length is kept at an initial value; when (when) And when the window length is more than 0.5s, the window length is reduced to 0.7 times of the initial value, and the initial window length is preset based on the IMU sampling frequency.
9. The fall prevention monitoring and early warning method based on remote monitoring as claimed in claim 8, wherein the preprocessing of the video data further comprises cluster optimization based on doppler power weights, and doppler power factors are introduced during the clustering process The clustering objective function is optimized as: , the doppler power weight for the point cloud data in the u-th video data, = , For the doppler power value of the point cloud data, And in the clustering process, the point cloud data with Doppler power higher than the average value is given higher weight, and the spatial characteristics of the point cloud data are reserved preferentially.
10. The fall-preventing monitoring and early-warning system based on remote monitoring is characterized by comprising a data acquisition module, a remote processing module, a fusion reasoning module and a hierarchical early-warning module, wherein the fall-preventing monitoring and early-warning system based on remote monitoring is used for realizing the fall-preventing monitoring and early-warning method based on remote monitoring as claimed in any one of claims 1 to 9; The data acquisition module comprises an IMU sensor, a 4D imaging radar sensor and RTLS positioning equipment, and is used for respectively acquiring motion characteristic data, environment video data and positioning data; the remote processing module is deployed at the cloud end and is used for executing preprocessing and feature extraction of the multi-source data; and the fusion reasoning module set is used for realizing fall risk judgment of multi-feature fusion.

Description

Fall-preventing monitoring and early-warning method and system based on remote monitoring Technical Field The application relates to the field of fall-preventing monitoring, in particular to a fall-preventing monitoring early warning method and system based on remote monitoring. Background With the increasing of global population aging and the increasing of the safety requirements of high-risk operation scenes, falling becomes an important hidden danger threatening the life safety of specific people. World health organization data show that falling is one of the main reasons of accidental injury of the old, and can cause serious consequences such as fracture, brain injury and the like, and high falling accidents at construction sites often cause serious injuries and deaths due to untimely monitoring. The current mainstream fall monitoring technology mainly comprises two types, namely a monitoring scheme based on wearable equipment, which has the problems of strong interference emotion of users, poor comfort level for long-term wearing and easiness in interference of limb actions, and high energy consumption in the data transmission process, and is difficult to meet long-time remote monitoring requirements, and a monitoring scheme based on visual images, which collects videos through a camera and analyzes human body gestures, which is easily influenced by illumination conditions and shielding conditions, has serious privacy leakage risks and is not suitable for private scenes such as families and wards. In addition, the prior art generally has the problem of response lag, most schemes can trigger early warning only after falling, early recognition of a pre-collision stage is difficult to realize, data processing depends on local equipment or cloud single nodes, and real-time performance and calculation force requirements cannot be considered. Therefore, it is necessary to provide a remote fall-preventing monitoring and early warning method and a corresponding system which have the advantages of non-contact performance, high accuracy, low false alarm rate and multi-scene suitability. Disclosure of Invention Based on this, it is necessary to provide a fall-preventing early warning method and system based on remote monitoring to effectively realize remote fall-preventing monitoring with non-contact, high accuracy, low false alarm rate and multiple scene suitability. The fall-preventing monitoring and early-warning method based on remote monitoring is characterized by comprising the following steps of: Acquiring multi-source data of a target object in real time, wherein the multi-source data comprises motion characteristic data, environment video data and positioning data; performing remote preprocessing and feature extraction on the acquired multi-source data, and generating an angular velocity root mean square sequence and an acceleration root mean square sequence based on the motion feature data; The system comprises a remote fusion reasoning model, a second decision unit, a first decision unit and a first analysis unit, wherein the fusion reasoning model is used for executing anti-falling monitoring early warning through the remote fusion reasoning model and comprises a two-stage decision unit, the first decision unit is used for calculating the slope angle of the angular speed root mean square sequence and triggering abnormal verification of an acceleration signal; If the local equipment and the cloud end judge that the falling risk or the falling occurs, starting acousto-optic early warning by the local equipment, and simultaneously pushing early warning information comprising falling time, position and risk level to a remote monitoring platform; If only a single level is judged to be a risk, starting a dynamic observation mode, encrypting data acquisition frequency and continuously monitoring characteristic change, wherein the single level is only a local or cloud fall risk or a fall occurs. The motion characteristic data can be acquired by an Inertial Measurement Unit (IMU) and comprises 3D gyroscope angular velocity data and 3D accelerometer acceleration data, the environment video data can be acquired by a 4D imaging radar sensor and output point cloud data comprising space coordinates, speed and Doppler power, and the positioning data can be acquired by a real-time positioning system (RTLS) and comprises space position coordinates of a target object; Performing remote preprocessing and feature extraction on the acquired multi-source data, performing dimension reduction on the motion feature data through Root Mean Square (RMS) calculation to generate an angular velocity root mean square sequence and an acceleration root mean square sequence, performing frame extraction, noise reduction, standardization and k-means clustering on the environment video data, and unifying data dimensions; The method comprises the steps of executing fall-prevention monitoring and early warning through a remote fusion reason