TW201535303A - An method and a system for detecting the off-bed behavior of the humans - Google Patents

Abstract

This invention discloses a method for detecting the off-bed behavior of the humans to solve the problem of the off-bed behavior is not able to early detecting. The method comprises a sensing step, an analysis step and a warning step. The sensing step is sensing an amount of a movement of a patient by a movement sensing module mounted on the patient and translating the amount of the movement into a movement data. The analysis step is inputting the movement data into a neural network with posture training to get a plurality of continuous human postures and determining weather the human postures satisfies the off-bed behavior of the humans to execute the warning, or not to re-execute the sensing step. The warning step is transmits a warning message to a monitoring end for sending out the warning message. This invention further discloses a system for detecting the off-bed behavior of the humans. Thus, it can actually resolve the said problem.

Description

Human body bed sensing method and system

The invention relates to a human body bed sensing method and system; in particular to a human body bed sensing method and system using an inverted transmission neural network.

In recent years, with the increasing number of children and aging, the issue of health care has gradually received attention. According to the statistical results, about 30% of elderly patients in the hospitalization process have experienced in-hospital falls, which has led to a worsening of the condition. After investigation, it is found that the fall usually occurs when the patient just walks off-bed. It happened, but the patient often did not know that he would fall after leaving the bed. Therefore, in order to prevent the fall from the bed, medical care workers need to know immediately whether the patient has left the bed in order to assist from the side.

In order to improve the fall in the hospital, the medical care home has introduced the human body bed sensing device, such as: image-type bed-sensing device, fence-type bed-sensing device or mattress-type bed-sensing device, in order to assist medical care workers to learn the disease. Suffering from information on leaving the bed can reduce the frequency of elderly patients falling in the hospital.

However, the conventional mattress-type bed-away sensing device sends a warning message (such as a bed, etc.) to the caregiver when the care recipient is out of bed, but in the process of the caregiver rushing to the care recipient's bed, The movement of the caregiver from the bed continues, so that when the caregiver arrives at the bedside, the caregiver often walks out of bed and falls, or has left the bed by himself; in addition, the conventional image-type bed-sensing device often requires manpower. Interpreted, and the sensing effect at night is not good; and the custom fence The type of bed-sensing device is prone to error. Therefore, it is obvious that the above-mentioned conventional human body bed-sensing device has limited warning function, and it is impossible to detect the behavior of leaving the bed early, and the improvement effect on the fall of the hospital is not good. Moreover, the statistical result of the fall from the bed can also be known that the medical care institution introduced the above. After knowing that the human body had left the bed to sense the equipment, the number of people falling in the hospital did not drop significantly.

In view of this, in the practical application of the human body bed-sensing device and method, in addition to the problem of "not being able to detect the problem of leaving the bed early", there are other doubts such as "poor sensing effect" and "prone to error", in actual use. There are many limitations and shortcomings, and there are inconveniences that need further improvement to improve its practicality.

The main object of the present invention is to provide a method for sensing a person's bed-awayness, which can immediately send a warning message when the subject is perceived to have a tendency to get out of bed.

A second object of the present invention is to provide a human body bed sensing system that immediately issues a warning message when the subject is perceived to have a tendency to get out of bed.

The present invention provides a method for sensing a human body leaving the bed, comprising: a sensing step of sensing a spatial movement amount of the care recipient by a motion sensing module provided by the care recipient, and converting the spatial movement amount by a servo end to a moving amount data; an analyzing step, the servo end inputting the moving amount data into a neural network of completed posture training, to obtain a plurality of continuous human body postures, and determining whether the human body is out of bed according to the human body postures Inertial action, if the determination is yes, perform an alerting step, if the determination is no, re-perform the sensing step; and in the warning step, the server sends a warning message to a monitoring terminal, and the monitoring terminal sends the warning message.

Preferably, the neural network is a reverse transmission type neural network.

The invention further provides a human body departure sensing system, comprising: a mobile sensing module for sensing a spatial position of a human body; a servo end coupled to the mobile sensing module, and analyzing whether the human body is based on the spatial position There is a trend of leaving the bed and an analysis result is generated; and a monitoring terminal is used to output the analysis result.

Preferably, the mobile sensing module is a device having a mobile sensing and wireless communication function.

Preferably, the server includes a first wireless access device, an attitude value search engine, an attitude determination server, an integrated service server, a wireless router, and a second wireless access device. The sensor is coupled to the motion sensing module and electrically connected to the attitude value search engine. The attitude value search engine, the attitude determination server and the integrated service server are connected to each other, and the integrated service server transmits the wireless The router and the second wireless access device are coupled to the monitoring terminal.

〔this invention〕

1‧‧‧Mobile sensing module

2‧‧‧Server

21‧‧‧First wireless access device

22‧‧‧ pose numerical search engine

23‧‧‧ attitude discriminating server

24‧‧‧Integrated Service Server

25‧‧‧Wireless Router

26‧‧‧Second wireless accessor

3‧‧‧Monitor

F‧‧‧ data format

K‧‧‧Sensing package

P‧‧‧ care recipients

R‧‧‧ class neural network

S1‧‧‧Sensing steps

S2‧‧‧ Analysis steps

S3‧‧‧ warning steps

1 is a system block diagram of an embodiment of a human body bed sensing method of the present invention.

2 is a flow chart showing the operation of the embodiment of the human body bed sensing method of the present invention.

Figure 3 is a schematic view of the human body coordinates and earth coordinates of the present invention.

Figure 4 is a schematic diagram of a neural network as used in the embodiment of the human body bed sensing method of the present invention.

Fig. 5 is a schematic view showing the human body leaving mode of the present invention.

The above and other objects, features and advantages of the present invention will become more <RTIgt; (off-bed) means the process by which the human body leaves the bed from the upper surface of the bed, as will be understood by those of ordinary skill in the art to which the present invention pertains.

The "coupled connection" as used throughout the present invention refers to the transfer of data between two devices by means of wired or wireless communication, such as following the same wired or wireless communication protocol, which is within the technical field of the present invention. Usually the knowledge person can understand.

The "Back-propagation neural network" (BPN) described in the full text of the present invention refers to a multi-layer perceptron (MLP) and error inversion algorithm. (Error Back Propagation, EBP) fusion neural network technology (see "Zhang Feizhang, Zhang Liqiu," Neural Network", Donghua Book Company, 2005"), in which the reverse transmission neural network will be in the learning phase. The error value generated at the output is transferred from the output layer back to the hidden layer, to the input layer, and the bond value between the networks is corrected to obtain a closer to the desired output result. In addition, the learning algorithm establishment (training) process of the inverse transfer type neural network includes: (1) setting network parameters; (2) setting the weighting matrix and the initial value of the partial weight vector by uniformly random random numbers; (3) calculating Hidden layer output; (4) set the output layer (Output layer) and hidden layer tolerance; (5) calculate the output layer and hidden layer gap; (6) whether the output layer and hidden layer gap is greater than The allowable gap is executed if it is established; (7) Calculate the output layer and hidden layer weighting matrix and the bias value correction amount; (8) Correct the output layer and hidden layer weight matrix and the partial weight; if the step is not satisfied (6) ), repeat (3) ~ (8) until the gap between the output layer and the hidden layer is within the allowable gap; (9) find the best mode (see "Yi Yicheng," neural network mode application and implementation ", Rulin Books Co., Ltd., 2004") is understood by those of ordinary skill in the art to which the present invention pertains.

Please refer to FIG. 1, which is a system block diagram of an embodiment of the human body bed sensing method of the present invention. The human body departure sensing system comprises: a mobile sensing module 1, a servo terminal 2 and a monitoring terminal 3, wherein the mobile sensing module 1 is a device having a mobile sensing and wireless communication function, such as: having a mobile Sensors (such as: Accelerometer, Accelerometer, G-sensor) smart phones, etc., to sense the spatial position of a human body, such as: three axes (X, Y, Z-axis) acceleration, etc.; the servo terminal 2 is a device having data processing and wireless communication functions, such as a computer having a wireless communication function, a server, or a device providing a cloud service. The mobile terminal 2 is coupled to the mobile sensing module 1 and the monitoring terminal 3 for executing a software program to analyze whether the human body has a tendency to leave the bed according to the spatial position, and transmit the analysis result to the The monitoring terminal 3; the monitoring terminal 3 is a device having data output and wireless communication functions, such as a notebook computer, a tablet computer or a line. A mobile phone or the like for outputting the analysis result, such as an analysis screen, a warning message, and/or a sound.

In this embodiment, the motion sensing module 1 is preferably fixed to a chest position of the care receiver P (as shown in FIG. 1) to detect a change in posture of the care recipient P; the servo end 2 The invention includes a first wireless access point 21, a sensor value assembly engine 22, an intelligent human body posture determine server 23, and an integrated service server (integrated) The service server 24, a wireless router 25 and a second wireless accessor 26, the first wireless accessor 21 is coupled to the mobile sensing module 1 and electrically connected to the attitude value search engine 22, the sensing signal of the mobile sensing module 1 is transmitted to the attitude value search engine 22, and the posture value search engine 22, the posture determination server 23, and the integrated service server 24 can communicate with each other through the Internet N. The gesture value search engine 22 is configured to collect the sensing signal and package the packet into a sensing packet, such as a packet including a packet number, a sensing module number, a packet type, a communication channel number, and a sensing data. , In the sensing recoverable conventional packet communication packet format, which is well known in the art that encapsulation system can be appreciated, this capacity is not repeated here.

The gesture discriminating server 23 is configured to parse the packet and analyze the human body posture represented by the sensing data by using a data mining technique, for example, analyzing the victim by using a reverse transmission neural network (BPN). Does P have a squat, squat, lie down, sit down to lie down, lie down to sit, turn right, turn right and then get up, turn left and turn left, then turn left and get up Posture, to determine whether the caregiver P has a tendency to get out of bed (eg, a continuous posture of the bed-away behavior), wherein the data analysis method can be used to analyze the algorithm, so that those skilled in the art can It is understood that the integrated service server 24 can pre-store information such as patient information and bed information, and is coupled to the monitoring terminal through the Internet N, the wireless router 25, and the second wireless accessor 26. 3, in order to send the caregiver P's bed information to the monitoring terminal 3 (such as: audio-visual equipment near the nursing staff) to issue an alarm when the care recipient P has a tendency to get out of bed The message, such as: 〝 indicates the willingness to leave the bed with different lights, or the broadcast xxx is about to leave the bed, so that the nursing staff can go to the bed of the care recipient P early, or use the talker to remind the care recipient P to stay out of bed. In order to reduce the number of occurrences of falls in the hospital, the integrated service server 24 can execute a data server to perform the above functions, which is well understood by those skilled in the art and will not be described herein.

Referring to FIG. 2, it is an operational flowchart of an embodiment of the method for sensing a human body leaving the bed according to the present invention, wherein the human body bed sensing method includes a sensing step S1, an analyzing step S2, and an alerting step S3. For a description of each, please refer to Figure 1 together.

In the sensing step S1, the motion sensing module provided in the care recipient senses the spatial movement amount of the care recipient, and the servo end converts the spatial movement amount into a movement amount data. In this embodiment, first, the motion sensing module 1 is fixed to the upper body of the care receiver P, so that the motion sensing module 1 continues (eg, every 10 seconds) to sense the caret P. The three-axis position, such as: using the X, Y, and Z components of the acceleration, and converting the X, Y, and Z components of the acceleration into the gravitational (g) value to calculate the angle between the human body coordinates and the earth coordinates, and The packet form is transmitted to the server terminal 2.

For example, the acceleration sensor has X, Y, and Z axes (orthogonal coordinates), and the sensing function is six directions. The same axis can be divided into two directions, using “+” and “- "Do the direction to distinguish. When the X axis and the Y axis are parallel to the earth, and the X axis and the Y axis output 0g without any external force, the Z axis will be affected by the gravitation of 1g; if the X axis and the Z axis are parallel to the earth, then The Y axis will be subjected to 1 g of gravity.

Please refer to FIG. 3 together, which is a schematic diagram of the human body coordinates and the earth coordinates of the present invention. Wherein, if the human body coordinates are {X', Y', Z'}, the earth coordinates are {X, Y, Z}, the angle between X and X' is α, and the angle between Y and Y' is β, Z. The angle with Z' is γ. Wherein, the value obtained by the acceleration sensor is an acceleration component of three axes of X, Y, and Z ( , , Therefore, before calculating the angles α, β, γ, we need to calculate the acceleration by using the Euclidean distance formula. , the calculation formula is as shown in formula (1): Next, the angles α, β, and γ of the human body coordinates and the earth coordinates are calculated as shown in the following formulas (2a), (2b), and (2c):

In this embodiment, the axis coordinates {X, Y, Z} are 0 degrees in the axial direction, and the angle angle or the human body coordinates can be known from the angles obtained by the above equations (2a), (2b), and (2c). {X', Y', Z'} can be used as the above-mentioned movement data, as a basis for subsequent evaluation of the human body posture. Taking the angle of the angle as an example, the angle range of the three axes can be normalized to 0~+90 and 0~ Between -90, use this definition to know the current space coordinates of the care recipient, but not limited to this.

In the analyzing step S2, the servo terminal 2 inputs the movement amount data into a neural network of completed posture training to obtain a plurality of continuous human body postures, and according to the human body postures, determine whether the inertia movement of the human body to the bed is satisfied. If the determination is YES, the above-described warning step S3 is performed, and if the determination is "NO", the above-described sensing step S1 is performed again. In this embodiment, as shown in FIG. 4, since each sensing packet K contains only part of the profile of the posture of the care recipient P, the server 2 can first collect the posture data in the packet. Such as: [T ₁ , X ₁ , Y ₁ , Z ₁ ], [T ₂ , X ₂ , Y ₂ , Z ₂ ], [T ₃ , X ₃ , Y ₃ , Z ₃ ], [T ₄ , X ₄ , Y ₄ , Z ₄ ], ..., [T _j , X _j , Y _j , Z _j ], until the servo terminal 2 judges that the collected posture data is sufficient to analyze a human body posture (eg, information contained in 40 packets) Or 100 pieces of data, etc., and then convert the posture data into a data format F required for the neural network to input a trained neural network R. For example, the Matlab software library is used to implement the inverse transfer neural network (BPN), and the data of the human body posture is input in advance to perform the weight training and adjustment of the neural network. After the training is completed, the human body represented by the posture data can be discriminated. Whether the posture belongs to the off-bed posture contained in a bed-away mode database, as shown in the following Table 1:

Thereby, the servo end 2 can judge whether or not it belongs to the bed-away mode by the posture of the care recipient. Please refer to FIG. 5, which is a schematic diagram of the human body bed-away mode of the present invention, wherein (a)-(f) respectively indicate "sitting on the edge of the bed (before the bed)" and "sitting on the feet (out of the bed)" , "Single-footed (out of bed)", "turning out of bed (out of bed)", "falling from bed (after leaving the bed)", "walking away from bed (after leaving the bed)" and other bed-away mode.

In addition, in order to avoid misjudging the posture of the human body, it is preferable to collect a plurality of continuously occurring posture data (for example, three posture data), so as to judge whether the care recipient is out of the bed by the inertial action when the human body is out of bed, and to judge the protection. The strength of the person's willingness to leave the bed, in order to issue a warning message early. For example, if the caregiver does not intend to leave the bed, the action on the bed is mainly lie down, 〝 left turn over and 〝 right turn over, and the person with mobility is less likely to fall asleep; in addition, when the guard Gesture The trend is a continuous movement such as a left turn to the back or a right turn and a right turn, indicating that the care recipient may be willing to leave the bed.

Moreover, when the care recipient gets up and then lie back on the bed, indicating that the care recipient has no intention to leave the bed, and does not need to issue a warning message (eg, xxx does not intend to leave the bed or display a green light number in the position of the bed); If you have not been lying back on the bed for a while, it means that the caregiver is likely to leave the bed, and can send further warning messages (such as: xxx is about to leave the bed or display a yellow light at the bed position); If the posture of the person is 〝站〞 (has left the bed), it may be normal walking (the angle of the Y-axis does not change much), or it may fall after 3~5 steps (the angle of the Y-axis changes greatly), and further warning messages are sent ( Such as: xxx has fallen or displayed a red light in its bed position, etc.). If the warning message is not required, the sensing step S1 is re-executed; if the warning message is to be sent, the warning step S3 is further performed.

In the warning step S3, the server sends a warning message to the monitoring terminal according to the human body posture satisfying the inertia action, and the monitoring terminal sends the warning message. In detail, if the server 2 senses that a certain insured person has a willingness to leave the bed, the warning message is generated and transmitted to the monitoring terminal 3, and the monitoring terminal 3 sends the warning message to let the nursing staff know that there is The position of the care recipient who wishes to leave the bed in order to provide timely assistance or to remind the care recipient with a talker. In addition, the servo end 2 can also generate a warning value according to the above-mentioned strength of the intention to leave the bed, such as: unintentional departure from the bed is 0, intentional departure from the bed 1, departure from the bed 2, normal walking to 3, having fallen to 4, etc., and the warning The value is transmitted to the monitoring terminal 3, and the warning value is outputted by the monitoring terminal 3, for example, the intensity of the willingness of all the insured persons is displayed in different colors for the caregiver to recognize, but not limited thereto.

For example, in order to confirm the correct judgment rate of each posture, it is divided into a squat, a squat, a lie, a lie down to a sitting squat, a squat to a lie, a squat and a squat, and a left turn over. After getting up and doing seven experiments, each pose gets more than 100 pieces of data. During the screening process, the sitting and the station are merged into one pose, and some invalid values are discarded, and then the data of each pose is divided. For the A and B groups, the data of the A group is used to train the neural network, and 50 pen values are input for each posture. The B group data is used for testing. The amount of data used by the two groups is as shown in the second list:

Therefore, the data of Group B is also pruned into 40 packets of data according to the pruning mode of Group A data, and then input into the neural network for classification. The results are shown in Table 3 below:

In the static action, the seat and the station are classified as "upper body upright", and the body inertia action is judged to distinguish the posture as a station or a seat. In the third table, the recognition rate of the static action is as high as 100%, and the dynamic action The average recognition rate of action is 95.3%. It belongs to a high recognition rate, and it can be seen that the training and weight adjustment of the neural network such as the present invention are correct, and the recognition rate is high and the occurrence of misjudgment can be avoided.

As another example, for the dynamic action, the original data of the B group is 100 packets. The action data is further divided into two groups, B1, each piece of data, 100 packets are divided by 40-40-20; B2, each piece of data, 100 packets are divided by 20-40-40, the number of divisions As shown in Table 4.

Therefore, each piece of information can obtain two 40 packets, including the static data part. To sit down and sit down, the action of the 100 packets of B1 should be "lie down, lie down to sit → sit"; In order to turn right up, the information should be “lie down → right turn up → sit”, therefore, in terms of lying down, according to the data splitting methods of B1 and B2, the classification of B1 should appear “lie” and “ Lying down to sit"; the classification of B2 should take the posture of "lying to sit" and "sit", the classification and quantity are shown in Table 5.

According to the above, the data of 100 packets is cut and the contents of the package are observed. It can be known that the action data is in the middle of 100 packets. After cutting, 40 packets may only contain most of the movements. After cutting, The information is not exactly as the above example is completely determined that 40 packets contain complete actions.

As shown in Table 6, almost 90% of the correctness is found in the identification of static actions. In addition, as shown in Table 7, when the action completion degree, that is, the action of lying down, the B1 can surely recognize "lie down, lie down to sit", or at B2, it can be surely recognized. "lie down Sitting → sitting on the integrity. Among them, in the completeness of the action, sitting and lying down will present a higher recognition rate because the sitting and lying movements are slower, so it has a better effect when sitting and lying down.

The main features of the human body bed sensing method and system embodiment of the present invention are as follows: First, the human body coordinate of the care recipient is sensed by the mobile sensing module, and is calculated by the server. The angle between the human body coordinates and the coordinates of the earth's earth; then, the servo end inputs the angle of the angle into the neural network of the above-mentioned neural network to obtain a plurality of continuous human body postures, and according to the postures of the human body, determine whether the inertia of the human body is removed from the bed. If the determination is "Yes", the above warning step is performed, and if the determination is "No", the sensing step is performed again; after that, the servo end transmits a warning message to the monitoring terminal according to the human body posture satisfying the inertial motion. The alert message is sent by the monitoring terminal. Thereby, the invention can effectively perceive the tendency of the care recipient to leave the bed, and when determining that the care recipient has the willingness to leave the bed, early issue a warning message to notify the nursing staff to perform relevant treatment to prevent the care recipient from falling out of bed and falling down. , to achieve the "effectively reduce the fall of the care recipient" effect.

While the invention has been described in connection with the preferred embodiments described above, it is not intended to limit the scope of the invention. The technical scope of the invention is protected, and therefore the scope of the invention is defined by the scope of the appended claims.

S1‧‧‧Sensing steps

S2‧‧‧ Analysis steps

S3‧‧‧ warning steps

Claims (5)

A human body bed sensing method includes: a sensing step of sensing a spatial movement amount of the care recipient by a motion sensing module provided by the care recipient, and converting the spatial movement amount to a movement amount by a servo end Data; an analysis step, the server inputs the movement amount data into a neural network of completed posture training, to obtain a plurality of continuous human body postures, and according to the human body postures, whether or not the inertia movement of the human body to the bed is satisfied is determined. If the determination is yes, an alarming step is performed. If the determination is no, the sensing step is performed again; and in the warning step, the server transmits a warning message to a monitoring terminal, and the monitoring terminal sends the warning message. The human body bed sensing method according to claim 1, wherein the neural network is a reverse transmission type neural network. A human body departure sensing system comprises: a mobile sensing module for sensing a spatial position of a human body; a servo end coupled to the mobile sensing module, and analyzing whether the human body has a tendency to leave the bed according to the spatial position And generating an analysis result; and a monitoring end for outputting the analysis result. The human body bed sensing system according to claim 3, wherein the mobile sensing module is a device having a mobile sensing and wireless communication function. The human body bed sensing system according to claim 3, wherein the server includes a first wireless access device, a pose value search engine, a gesture determination server, an integrated service server, a wireless router, and a second wireless access device, the first wireless access device is coupled to the motion sensing module, and electrically connected to the attitude value search engine, the attitude value search engine, the attitude determination server, and the integrated service servo Connected to each other, the integrated service server accesses the wireless router and the second wireless The device is coupled to the monitoring terminal.