CN-122004844-A - Method and system for preventing dementia patient from falling down

CN122004844ACN 122004844 ACN122004844 ACN 122004844ACN-122004844-A

Abstract

The invention relates to the technical field of medical assistance, in particular to a method and a system for monitoring fall prevention of a dementia patient. A monitoring system for preventing a dementia patient from falling comprises a dementia patient monitoring module, a motion parameter data construction module, a motion control analysis module and a falling intervention prevention module. The invention continuously collects video data of the space where a dementia patient is located, extracts the motion parameters of key nodes of trunk and lower limbs by combining skeleton topology and time sequence constraint, constructs a motion control constraint space reflecting the cooperative motion relation of human bodies, further quantifies the effective degree of freedom of motion control, realizes the advanced identification of the state of human body stability decline before falling, and timely triggers intervention measures when detecting that the effective degree of freedom of motion control is lower than a instability threshold.

Inventors

ZHONG QIONG
HU JIANHUA
LAI YUKANG
WU XIUYAN
LI HUI

Assignees

赣州市第三人民医院

Dates

Publication Date: 20260512
Application Date: 20260311

Claims (7)

1. A method for fall prevention monitoring of a dementia patient, comprising: sampling video data based on a preset time length to obtain a video frame sequence; processing the video frame sequence based on skeleton topology and time sequence constraint to obtain motion parameter data of a motion key node corresponding to a dementia patient; Determining a motion control constraint space based on the motion parameter data, and determining a motion control effective degree of freedom in the motion control constraint space according to the motion parameter data; And comparing the motion control effective degree of freedom with the instability threshold, if the motion control effective degree of freedom is smaller than the instability threshold, taking intervention measures for the corresponding dementia patient, and if the motion control effective degree of freedom is not smaller than the instability threshold, not operating.
2. The method for monitoring fall prevention of a dementia patient according to claim 1, wherein the video frame sequence is processed based on skeleton topology and time sequence constraint to obtain motion parameter data of a motion key node corresponding to the dementia patient, and the method specifically comprises the following steps: Detecting human body areas of all video frames in the video frame sequence, forming a human body area image sequence from all obtained human body area images, and marking the human body area images through time stamps corresponding to the video frames; For each human body region image in the human body region image sequence, performing the following operation, sending the human body region image into a key part labeling network for processing to obtain labeling information corresponding to a plurality of human body structure regions in the human body region image, wherein the labeling information comprises region center coordinates and corresponding labeling labels, and connecting motion key nodes based on the labeling labels corresponding to the human body structure regions under the constraint of a skeleton topology rule to construct a skeleton topology graph; And tracking the motion key nodes based on the labeling information corresponding to the human body region image and the skeleton topological graph, and constructing motion parameter data of the motion key nodes corresponding to the dementia patient.
3. The method for fall prevention monitoring of a dementia patient according to claim 2, wherein the human body region detection is performed on all video frames in the video frame sequence, and the obtained human body region images are formed into a human body region image sequence, comprising the steps of: Executing the following contents aiming at each pixel of the video frame, acquiring pixel values of the pixels in all the video frames, and fitting a Gaussian distribution model corresponding to one pixel aiming at all the pixel values, wherein the Gaussian distribution model comprises an observation mean value and an observation variance; Traversing each pixel of the video frame in the video frame sequence, matching the pixel value corresponding to the pixel with a Gaussian distribution model corresponding to the pixel, wherein the matching mode is to compare the absolute value of the difference value between the pixel value corresponding to the pixel and the observed mean value with the product of the observed variance and the empirical coefficient, if the absolute value of the difference value between the pixel value corresponding to the pixel and the observed mean value is smaller than the product of the observed variance and the empirical coefficient, the matching is considered successful, if the matching of the pixel value corresponding to the pixel and the Gaussian distribution model corresponding to the pixel is successful, the pixel is marked as a background pixel, if the matching of the pixel value corresponding to the pixel and the Gaussian distribution model corresponding to the pixel is unsuccessful, the pixel is marked as a non-background pixel, then carrying out connected domain analysis on all the non-background pixels in the video frame, marking the rectangular image corresponding to the largest connected domain as a human body region image, and setting the pixel value of the pixel which does not belong to the largest connected domain in the human body region image as 0; And all the obtained human body region images are formed into a human body region image sequence.
4. The method for monitoring fall prevention of a dementia patient according to claim 3, wherein the method is characterized in that the motion key nodes are tracked based on the labeling information corresponding to the human body region image and the skeleton topological graph, and the motion parameter data of the motion key nodes corresponding to the dementia patient are constructed, and specifically comprises the following steps: Performing union processing on labeling labels corresponding to all the skeleton topological graphs to obtain a labeling label item set; Traversing label tags in a label item set, executing the following content aiming at each label tag, traversing a human body region image, inquiring in label information corresponding to a current human body region image based on the label tag, selecting a region center coordinate in label information corresponding to the human body region image as a positioning coordinate if the label tag exists in label information corresponding to the human body region image, labeling the label tag through a time stamp corresponding to a human body region image sequence, if the label tag does not exist in label information corresponding to the human body region image, inquiring in label information corresponding to a human body region image adjacent to the current human body region image based on the label tag, marking the human body region image with the label tag as a positioning human body region image based on the label tag, inquiring from a skeleton topological graph corresponding to the positioning human body region image if the label tag exists in label information corresponding to the human body region image, marking the label tag corresponding to the motion key node adjacent to the label tag in the skeleton topological graph corresponding to the human body region image, marking the motion key node as a positioning motion key node in label information corresponding to the label region image, fitting the label coordinate of the label coordinate in the label coordinate of the human body region corresponding to the human body region image to be fitted to the coordinate of the label region to obtain a difference value, fitting the coordinate of the label region to the human body region to be fitted to the coordinate of the label region to the coordinate of the coordinate to be fitted, the positioning coordinates or the fitting coordinates are arranged and spliced according to the sequence numbers in the human body region image sequence to form a labeling tag data sequence set, wherein the labeling tag data sequence set comprises labeling tag coordinates arranged according to the sequence numbers in the human body region image sequence; For each labeling tag data sequence set, marking the distance between adjacent labeling tag coordinates as motion displacement, marking the time stamp corresponding to the labeling tag coordinate positioned at the later in the adjacent labeling tag coordinates, arranging and splicing all the motion displacement according to the time stamp to form a motion displacement vector, marking the interval between the labeling tag coordinates and the time stamp corresponding to the first labeling tag coordinate in the labeling tag data sequence set as interval time corresponding to the labeling tag coordinate, taking the interval time as an independent variable, taking the motion displacement as a dependent variable, fitting a curve function, performing first-order differentiation, taking data corresponding to the interval time as motion velocity, performing second-order differentiation, taking data corresponding to the interval time as motion acceleration, arranging and splicing all the motion velocity according to the interval time to form a motion velocity vector, and arranging and splicing all the motion acceleration according to the interval time to form a motion acceleration vector; and splicing the motion displacement vectors, the motion velocity vectors and the motion acceleration vectors corresponding to the labeling data sequence sets from top to bottom to form local motion parameter data, forming the local motion parameter data corresponding to all the labeling data sequence sets into overall motion parameter data, wherein the overall motion parameter data is actually matrix data, and then recording the transpose of the overall motion parameter data as the motion parameter data.
5. The method for fall prevention monitoring of dementia patients according to claim 4, wherein the motion control constraint space is determined based on the motion parameter data, and the motion control effective degree of freedom is determined in the motion control constraint space according to the motion parameter data, specifically comprising the following: constructing a covariance matrix corresponding to the motion parameter data, decomposing the characteristic value of the covariance matrix corresponding to the motion parameter data, and marking the ratio of the obtained maximum characteristic value to the sum of all the characteristic values as the change consistency; each line in the motion parameter data is marked as a motion parameter vector, the product of the motion parameter vector and a motion control constraint matrix is marked as fitting change consistency, mean square errors between all fitting change consistency and change consistency are calculated and marked as mapping errors, the motion control constraint matrix is subjected to iterative optimization through a least square method in a direction of minimizing the mapping errors until a termination condition is met, the motion control constraint matrix is output, and the transposition of the motion control constraint matrix is the motion control constraint space; singular value decomposition is carried out on the motion control constraint space, and the number of non-zero singular values is the effective degree of freedom of motion control.
6. The method for fall prevention monitoring of dementia patients according to claim 5, wherein training is performed for critical part labeling network, specifically comprising the following steps: And obtaining a plurality of key part labeling samples, labeling the key part labeling samples through labeling information, forming a key part labeling set by all labeled key part labeling samples, and training a key part labeling network through the key part labeling set.
7. A dementia patient fall prevention monitoring system, characterized in that the system applies a dementia patient fall prevention monitoring method as claimed in any one of the preceding claims 1-6, comprising: the dementia patient monitoring module is used for sampling video data based on a preset time length to obtain a video frame sequence; The motion parameter data construction module is used for processing the video frame sequence based on skeleton topology and time sequence constraint to obtain motion parameter data of the motion key nodes corresponding to the dementia patient; The motion control analysis module is used for determining a motion control constraint space based on the motion parameter data and determining a motion control effective degree of freedom in the motion control constraint space according to the motion parameter data; and the anti-falling intervention module is used for comparing the motion control effective degree of freedom with the instability threshold, taking intervention measures for the corresponding dementia patient if the motion control effective degree of freedom is smaller than the instability threshold, and having no operation if the motion control effective degree of freedom is not smaller than the instability threshold.

Description

Method and system for preventing dementia patient from falling down Technical Field The invention relates to the technical field of medical assistance, in particular to a method and a system for monitoring fall prevention of a dementia patient. Background The existing fall prevention technology for the elderly, particularly dementia patients, is mainly realized by means of wearable sensors, pressure sensing devices or post-video playback analysis and the like, wherein the wearable equipment is required to be worn by the patients for a long time, monitoring is easily interrupted due to poor compliance, misoperation or loss of the dementia patients, certain interference exists on daily activities of the patients, the scheme based on environment sensing or simple video analysis is mainly focused on recognition and alarm after falling, judgment is usually carried out through the appearance characteristics such as sudden change of human body posture, sudden increase of speed or contact with the ground, the gradual falling process of human body motion control capacity is difficult to capture in time, and the problems of early warning lag and high false alarm rate exist. Disclosure of Invention The method comprises the steps of continuously collecting video data in a ward, a household or a sanatorium and the like, sampling and graying the video frames based on a preset time window to achieve stable acquisition of the motion state of a dementia patient in a short time, identifying, connecting and tracking motion key nodes such as a trunk, a hip joint, a knee joint and an ankle joint in a video frame sequence through introducing skeleton topological rules and time sequence constraints, constructing high-dimensional motion parameter data comprising displacement, speed and acceleration, accurately representing the actual motion behavior of the dementia patient in a continuous time period, constructing a covariance matrix based on the motion parameter data, extracting a change consistency index, combining least square iterative optimization to obtain a motion control constraint space, quantifying the effective degree of freedom of the motion stability of the human body in the current state through singular value decomposition, identifying the motion stability reduction of the patient before the patient falls, and triggering voice prompt or manual intervention in time through comparing the motion control effective degree of freedom with a destabilization threshold. The invention provides a method for preventing a dementia patient from falling down, which comprises the following steps: sampling video data based on a preset time length to obtain a video frame sequence; processing the video frame sequence based on skeleton topology and time sequence constraint to obtain motion parameter data of a motion key node corresponding to a dementia patient; Determining a motion control constraint space based on the motion parameter data, and determining a motion control effective degree of freedom in the motion control constraint space according to the motion parameter data; And comparing the motion control effective degree of freedom with the instability threshold, if the motion control effective degree of freedom is smaller than the instability threshold, taking intervention measures for the corresponding dementia patient, and if the motion control effective degree of freedom is not smaller than the instability threshold, not operating. As a preferred aspect, the processing of the video frame sequence based on the skeleton topology and the time sequence constraint, to obtain the motion parameter data of the motion key node corresponding to the dementia patient, specifically includes the following steps: Detecting human body areas of all video frames in the video frame sequence, forming a human body area image sequence from all obtained human body area images, and marking the human body area images through time stamps corresponding to the video frames; For each human body region image in the human body region image sequence, performing the following operation, sending the human body region image into a key part labeling network for processing to obtain labeling information corresponding to a plurality of human body structure regions in the human body region image, wherein the labeling information comprises region center coordinates and corresponding labeling labels, and connecting motion key nodes based on the labeling labels corresponding to the human body structure regions under the constraint of a skeleton topology rule to construct a skeleton topology graph; And tracking the motion key nodes based on the labeling information corresponding to the human body region image and the skeleton topological graph, and constructing motion parameter data of the motion key nodes corresponding to the dementia patient. As a preferred aspect, the human body region detection is performed on all video frames in the video frame sequence, and all