Fall position recognition method and companion robot

Abstract

The application discloses a fall position recognition method and a companion robot, a human body posture angle and angular velocity collected through a VR head-mounted device are used for early warning of a falling behavior, the falling behavior of a VR experimenter can be quickly recognized, then human head trajectory prediction is carried out based on an LSTM coding and decoding model, the model has high prediction accuracy and can be used for path planning of the companion robot.

Description

Technical Field

[0001] This invention relates to the field of fall posture recognition, and in particular to a fall location recognition method and a companion robot. Background Technology

[0002] VR users are prone to falls in daily life for various reasons, such as: encountering uneven ground or obstacles while using VR devices, and the virtual environment not displaying the flatness of the obstacles and the actual ground, leading to falls; or due to poor quality or stability of the VR device, users may lose their balance and fall when moving or turning; or because VR experiences may require users to move or make quick movements, if users have poor balance or insufficient perception of the virtual environment, they may fall.

[0003] If no one notices and assists in time, it could have serious consequences. There have been reports of a 44-year-old Moscow resident who fell on a glass table while using a VR headset and died from cuts and blood loss. While existing monitoring systems or smart devices can detect falls to some extent, they cannot accurately assess the faller's condition, nor can they protect both the person and the VR device after a fall has occurred. Therefore, developing a system that can accurately identify falls and automatically provide assistance is of great importance. Summary of the Invention

[0004] The purpose of this invention is to provide a fall location recognition method and a companion robot, which can accurately and quickly identify the fall behavior and head landing position of a VR user, and protect the user by moving to the head landing position with the companion robot.

[0005] A method for identifying the location of a fall includes the following steps:

[0006] S1. Train the LSTM neural network;

[0007] Data collection: Collect time series of position coordinate data of VR head-mounted devices. The position coordinate data of each time step in the time series includes the coordinates (xi, yi, zi) of the VR head-mounted device in the navigation coordinate system, where i is the time step number, x is the coordinate of due east, y is the coordinate of due north, and z is the coordinate of the direction perpendicular to the horizontal plane.

[0008] Data preparation: The position coordinate data of each time step in the time series is standardized according to the values ​​of the three coordinate axes of the navigation coordinate system, so that the mean of the position coordinate data of each time step in the time series is 0 and the variance is 1.

[0009] Dataset partitioning: The standardized time series is divided into an input sequence and a target sequence;

[0010] LSTM Neural Network Training: Initialize the LSTM model parameters, propagate the input sequence forward through the LSTM model to obtain the predicted sequence, calculate the loss function based on the predicted sequence and the target sequence, update the weight parameters of the LSTM model based on the calculated loss function gradient through the backpropagation algorithm, and repeat the training process until the preset number of training times is reached or until convergence is achieved.

[0011] S2. Using the ground as a reference, establish a navigation coordinate system in the VR experience space and collect real-time data of the VR headset when the human wears the VR headset, including attitude angle, acceleration, and position coordinate data in the navigation coordinate system;

[0012] S3. Input the attitude angle and acceleration into the fall prediction model and extract the real-time data that triggers the fall conditions;

[0013] S4. Using the sliding window method, extract the position coordinate data corresponding to the time steps within the sliding window range from the real-time data extracted in step S3 to form a sample time series, wherein the time steps within the sliding window range include the time steps of the real-time data extracted in S3;

[0014] S5. After standardizing the time series of the S4 samples, input them into the trained LSTM neural network, and perform inverse standardization on the output to obtain the prediction results of the VR head-mounted device's position coordinate data.

[0015] The standardization process specifically includes the following steps:

[0016] Calculate the mean and variance of the position coordinate data for all time steps on the three coordinate axes of the navigation coordinate system.

[0017] The following formula is used to convert the position coordinate data of each time step in the time series into standardized values ​​corresponding to the three coordinate axes of the navigation coordinate system:

[0018] Standardized value = (Original value - Mean) / Variance;

[0019] The original value is the value corresponding to the selected coordinate axis in the navigation coordinate system for the position coordinate data of each time step. The mean is the average value of the position coordinate data of all time steps corresponding to the selected coordinate axis in the navigation coordinate system. The variance is the variance of the position coordinate data of all time steps corresponding to the selected coordinate axis in the navigation coordinate system.

[0020] Preferably, when partitioning the dataset, a sliding window method is used, where the standardized time series within a time window is used as the input sequence, and the time series at the next time step of the input sequence is used as the target sequence.

[0021] Furthermore, when partitioning the dataset, a sliding window method is used to divide the standardized time series within a time window into three consecutive time steps. The time series of the first two time steps are used as the input sequence, and the time series of the last time step is used as the target sequence.

[0022] In step S2, the attitude angles include the pitch angle, roll angle, and yaw angle. The yaw angle represents the offset angle with respect to the Y-axis; the pitch angle represents the offset angle with respect to the X-axis; and the roll angle represents the deflection angle with respect to the Z-axis. The X-axis is the axis perpendicular to the left and right sides of the human body, the Z-axis is the axis perpendicular to the front and back sides of the human body, and the Y-axis is perpendicular to the X-axis and Z-axis. The accelerations are the accelerations AccX, AccY, and AccZ along the X, Y, and Z axes, respectively.

[0023] The fall prediction model in step S3 includes an attitude angle differentiation unit and an acceleration differentiation unit. The fall prediction model performs the following steps to determine whether a fall condition is triggered:

[0024] After the posture angle is input into the posture angle differentiation unit, the formula (1) can be used to distinguish the forward bending and backward falling of the human body from other postures, and obtain sqrtRP;

[0025] sqrtRP calculates RPP_delta according to formula (2) and compares it with the set RPP_delta threshold, which can effectively extract the bending and falling movements of the human body in the lateral and backward directions.

[0026]

[0027] RPP_delta=fabs(SqrtRP-yaw) (2)

[0028] When the RPP_delta angle is greater than 30°, the acceleration will be input into the acceleration differentiation unit.

[0029] When the yaw angle drops to no more than 70°, the acceleration differentiation unit first calculates the combined acceleration of the Z-axis and Y-axis according to formula (3): SqrtAccYZ, based on the acceleration in the Z-axis direction and the acceleration in the Y-axis direction. Then, it calculates ACCYZZ_delta based on SqrtAccYZ, the acceleration in the Z-axis direction and formula (4), and then compares it with the normal threshold and the fall threshold to determine whether the human body has fallen.

[0030]

[0031] AccYZZ_delta=fabs(SqrtAccYZ-AccZ) (4)

[0032] Observe the change in ACCYZZ_delta angle. When ACCYZZ_delta is less than 40, it is judged as a fall, triggering the fall condition.

[0033] A companion robot travels to the predicted location coordinates in step S5 to provide protection. The companion robot carries a memory foam pad to absorb impact and energy, reducing the impact force on the head during a fall.

[0034] This invention provides a method for predicting human head trajectory based on an LSTM encoding and decoding model. The model has high prediction accuracy and can be used for path planning of companion robots. This invention also provides a method for early warning of fall behavior, which can quickly identify the fall behavior of VR users. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of an LSTM neural network.

[0036] Figure 2 This is a schematic diagram of human posture angles;

[0037] Figure 3 A schematic diagram of the fall prediction process for a fall prediction model;

[0038] Figure 4 A flowchart for path planning for a companion robot. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, and not all embodiments. The components described and illustrated in the accompanying drawings can typically be arranged and designed in various different configurations.

[0040] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0041] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0042] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0043] Example 1

[0044] This application proposes a method for identifying the location of a fall, as detailed below:

[0045] S1. Training the LSTM neural network

[0046] Data collection: Collect time series of position coordinate data of VR head-mounted devices. The position coordinate data of each time step in the time series includes the coordinates (xi, yi, zi) of the VR head-mounted device in the navigation coordinate system, where i is the time step number, x is the coordinate of due east, y is the coordinate of due north, and z is the coordinate of the direction perpendicular to the horizontal plane.

[0047] Data preparation: The position coordinate data of each time step in the time series is standardized according to the values ​​of the three coordinate axes of the navigation coordinate system, so that the mean of the position coordinate data of each time step in the time series is 0 and the variance is 1.

[0048] Dataset partitioning: The standardized time series is divided into an input sequence and a target sequence;

[0049] LSTM Neural Network Training: Initialize the LSTM model parameters, propagate the input sequence forward through the LSTM model to obtain the predicted sequence, calculate the loss function based on the predicted sequence and the target sequence, update the weight parameters of the LSTM model based on the calculated loss function gradient through the backpropagation algorithm, and repeat the training process until the preset number of training iterations is reached or until convergence is achieved.

[0050] S2. Using the ground as a reference, establish a navigation coordinate system in the VR experience space and collect real-time data of the VR headset when the human wears the VR headset, including attitude angle, acceleration, and position coordinate data in the navigation coordinate system;

[0051] S3. Input the attitude angle and acceleration into the fall prediction model and extract the real-time data that triggers the fall conditions;

[0052] S4. Using the sliding window method, extract the position coordinate data corresponding to the time steps within the sliding window range from the real-time data extracted in step S3 to form a sample time series, wherein the time steps within the sliding window range include the time steps of the real-time data extracted in S3;

[0053] S5. After standardizing the time series of the S4 sample, input it into the trained LSTM neural network, and perform inverse standardization on the output to obtain the prediction result of the VR head-mounted device's position coordinate data.

[0054] Specifically, the position coordinate data of the VR headset is obtained by a radar sensor installed on the headset. The antenna in the radar sensor emits electromagnetic waves, which propagate through space and are reflected, refracted, or scattered when they encounter objects. The receiver in the radar sensor receives the signal reflected back from the target. The received signal undergoes amplification, filtering, and modulation to extract relevant target information. By measuring the time difference between the transmitted and received signals, the radar sensor can calculate the target's distance.

[0055] Typically, radar sensors use a navigation coordinate system referenced to the local ground, and targets in the VR experience space have their positions defined relative to this coordinate system. Therefore, the position coordinates of the VR headset in the VR experience space can be obtained as (xi, yi, zi).

[0056] Preferably, the standardization process specifically includes:

[0057] Calculate the mean and variance of the position coordinate data for all time steps on the three coordinate axes of the navigation coordinate system.

[0058] The following formula is used to convert the position coordinate data of each time step in the time series into standardized values ​​corresponding to the three coordinate axes of the navigation coordinate system:

[0059] Standardized value = (Original value - Mean) / Variance;

[0060] The original value is the value corresponding to the selected coordinate axis in the navigation coordinate system for the position coordinate data of each time step. The mean is the average value of the position coordinate data of all time steps corresponding to the selected coordinate axis in the navigation coordinate system. The variance is the variance of the position coordinate data of all time steps corresponding to the selected coordinate axis in the navigation coordinate system.

[0061] Through the above standardization process, the mean of the standardized values ​​is 0 and the variance is 1.

[0062] Inverse standardization is used to calculate the true value based on the obtained standardized prediction results.

[0063] True value = Standardized prediction result * Variance + Mean

[0064] In step S1 of this embodiment, the input sequence is a standardized historical time series used as model input, and the target sequence is the standardized time series to be predicted for the next time step.

[0065] Specifically, the sliding window method is used, where the standardized time series within a time window is used as the input sequence, and the time series at the next time step of the input sequence is used as the target sequence.

[0066] Furthermore, when using the sliding window method, the standardized time series within a time window is divided into three consecutive time steps. The time series of the first two time steps are used as the input sequence, and the time series of the last time step is used as the target sequence.

[0067] Assume the time series after standardization is as follows:

[0068] t1:10,20,30,40;

[0069] t2:15,25,35,45;

[0070] t3:20,30,40,50;

[0071] t4:25,35,45,55;

[0072] Using the sliding window method, these data are transformed into input and target sequences:

[0073] Assuming the window size is 2, meaning each input sequence sample contains data from 2 time steps:

[0074] Input: [10,20,30,40], [15,25,35,45], Output: [20,30,40,50]

[0075] Input: [15,25,35,45], [20,30,40,50], Output: [25,35,45,55]

[0076] The sliding window method can divide continuous position data from a VR headset into an input window and a corresponding output.

[0077] In step S1, the LSTM neural network is a special type of RNN (Recurrent Neural Network), such as... Figure 1 The LSTM network structure inputs the position coordinates of the VR head-mounted device at time t into the network in chronological order. In the LSTM unit, the output of the current state needs to be calculated based on the state of the memory cell and the input, and the information of the memory cell is updated and passed to the next hidden layer unit. Finally, the position coordinate information at time t+1 is output.

[0078] In training the trajectory prediction model for VR headsets, it's necessary to predict the long-term trajectory based on previous VR headset trajectory data. This requires continuously iterating the LSTM network structure, using the output at time t+1 as the input at time t+2, and repeating this process to predict long-term VR headset trajectory information. Experiments were conducted using VR headset position coordinates with various step sizes as network inputs. During network training, the network evaluates and optimizes the predicted output based on the known VR headset trajectory information. The average 3D spatial distance between the predicted and actual values ​​is calculated as the training objective, and the mean squared error between the predicted and actual values ​​is used as the loss function. Backpropagation is employed to update the LSTM model's weights based on the calculated loss function gradient. This training process is repeated until the preset number of training iterations is reached or until convergence.

[0079] In step S2 of this embodiment, the VR headset uses the InvenSense MPU6050 chip, which integrates a 3-axis gyroscope and a 3-axis accelerometer, to detect human posture and gravitational acceleration. To better observe acceleration changes, the acceleration in the three directions is analyzed separately, establishing X, Y, and Z three-phase coordinate axes. For convenient analysis of posture angles, the MPU6050 chip is used to collect human data; it has its own three-axis coordinate system and can measure posture angles, namely the pitch angle, roll angle, and yaw angle. Figure 2As shown, the yaw angle represents the offset angle with respect to the Y-axis; the pitch angle represents the offset angle with respect to the X-axis; and the roll angle represents the deflection angle with respect to the Z-axis. The X-axis is perpendicular to the left and right sides of the human body, the Z-axis is perpendicular to the front and back sides of the human body, and the Y-axis is perpendicular to both the X and Z axes. This attitude coordinate system is established by mounting the MPU6050 chip on the VR headset. Accelerations are defined as AccX, AccY, and AccZ along the X, Y, and Z axes, respectively.

[0080] The fall prediction model in step S3 includes attitude angle differentiation units and acceleration differentiation units. The fall prediction model is executed. Figure 3 The judgment process is shown below.

[0081] After the posture angle is input into the posture angle differentiation unit, the formula (1) can be used to distinguish the forward bending and backward falling of the human body from other postures, and obtain sqrtRP;

[0082] sqrtRP calculates RPP_delta according to formula (2) and compares it with the set RPP_delta threshold, which can effectively extract the bending and falling movements of the human body in the lateral and backward directions.

[0083]

[0084] RPP_delta=fabs(SqrtRP-yaw) (2)

[0085] Under normal standing conditions, the RPP_delta angle does not exceed 30°. However, when bending over or falling, the RPP_delta angle exceeds 30°. Therefore, the RPP_delta threshold angle can be set to 30°. When the RPP_delta angle is less than 30°, it is considered a normal human activity state. When the RPP_delta angle is greater than 30°, the acceleration data is input into the acceleration differentiation unit for further discussion and judgment on the correlation between gravity and acceleration.

[0086] Since bending over and falling down involve the same change in posture, it is necessary to distinguish between normal bending over and falling postures based on the intensity of their movements. The intensity of the movements needs to be determined using gravitational acceleration.

[0087] When falling or leaning backward, the yaw angle will decrease. When the yaw angle decreases to the threshold of 70°, the acceleration discrimination unit first calculates the combined acceleration of the Z-axis and Y-axis according to formula (3): SqrtAccYZ. Then, based on SqrtAccYZ, the acceleration in the Z-axis direction and formula (4), ACCYZZ_delta is calculated and compared with the normal threshold and the fall threshold to determine whether the human body has fallen.

[0088]

[0089] AccY and AccZ represent the accelerations in the Y and Z directions, respectively. The Z-axis acceleration can indicate whether there is a violent backward movement, whether there is weightlessness, or whether there is a violent impact with the ground or seat. At the same time, the downward acceleration in the Y-axis direction will also change significantly during the process of falling backward.

[0090] AccYZZ_delta=fabs(SqrtAccYZ-AccZ) (4)

[0091] By observing the changes in the ACCYZZ_delta angle, the ACCYZZ_delta amplitude is around 80 when leaning back or lying down, while it is around 40 when a fall occurs. Therefore, we can set the normal threshold for ACCYZZ_delta to 80. If ACCYZZ_delta is greater than 80, it is judged as leaning back or lying down. We can set the fall threshold for ACCYZZ_delta to 40. If ACCYZZ_delta is less than 40, it is judged as a fall, thus further distinguishing normal movements such as leaning back from falling movements.

[0092] Example 2

[0093] After predicting the location where a person will fall, this embodiment uses a companion robot to travel to the predicted location coordinates in step S5 to provide protection.

[0094] A companion robot typically consists of the following components: chassis, wheels, battery, DC motor, DC motor driver module, Bluetooth communication module, core controller module, sensors, and memory foam padding. Memory foam has excellent energy absorption properties, effectively absorbing impact and energy, reducing the impact force during falls; it provides good shock absorption, gradually adapts to the shape of the head, and slows the propagation of impact; it can provide personalized support according to the shape and weight of the head, thus providing better support during falls.

[0095] The companion robot uses a path planning algorithm (A* algorithm) to calculate the path from the robot's current position to the target position.

[0096] The A* algorithm uses heuristic functions to search for paths and is suitable for path planning where global environment information is known. During the algorithm's execution, two lists are generated: an open list and a close list. The open list contains nodes that have been visited but not selected as the next node, while the close list contains nodes that have not been visited or have been visited but not expanded.

[0097] The A* algorithm performs path planning using a cost estimation function, expanding from the initial position to nearby child nodes. The cost estimation function selects the node with the lowest cost based on the distance cost of the child nodes as the next parent node, repeating this process until path planning is complete and the final path is generated.

[0098] The general expression for a heuristic function is:

[0099] f(n) = g(n) + h(n)

[0100] In the formula, f(n) represents the cost estimate of reaching the target point from the starting grid point through any node n, g(n) is the actual distance assessment from the starting point to grid n, and h(n) represents the distance assessment of the optimal route from grid n to the destination.

[0101] The path planning process is as follows Figure 4 As shown:

[0102] (1) Set up Openlist and Closelist. Openlist is used to store nodes that have been searched, and Closelist is used to store nodes that have not been explored.

[0103] (2) Set Nx as the initial grid point and Ny as the specified target point. Place the initial grid point in the Openlist. (3) Visit the nodes near the initial grid point, place non-obstacle points in the Openlist, calculate their estimated distances using heuristic functions, and place the initial grid point in the Closelist.

[0104] (4) Place the next expansion point N in the Closelist according to the optimal selection strategy. If N is the specified target grid, the path planning ends and the algorithm runs successfully. Otherwise, proceed to the following steps.

[0105] (5) Continue to implement the strategy of accessing its surrounding nodes for grid N. For surrounding child nodes that are not in the Closelist, add them to the Openlist and calculate the cost.

[0106] (6) Return to step (4) until the target point is reached.

[0107] (7) The algorithm runs successfully. Save the map and extended nodes. Backtracking the connection is the final path.

[0108] The location of the companion robot in the experience space can be determined by installing radar in the corner of the room and establishing a navigation coordinate system.

Claims

1. A method for recognizing the location of a fall, characterized in that, Includes the following steps: S1. Train the LSTM neural network; Data collection: Time series of location coordinate data of VR headset. The location coordinate data of each time step in the time series includes the coordinates (xi, yi, zi) of VR headset in the navigation coordinate system, where i is the time step number, x is the coordinate of due east, y is the coordinate of due north, and z is the coordinate of the direction perpendicular to the horizontal plane. Data preparation: The position coordinate data of each time step in the time series is standardized according to the values ​​of the three coordinate axes of the navigation coordinate system, so that the mean of the position coordinate data of each time step in the time series is 0 and the variance is 1. Dataset partitioning: The standardized time series is divided into an input sequence and a target sequence; LSTM Neural Network Training: Initialize the LSTM model parameters, propagate the input sequence forward through the LSTM model to obtain the predicted sequence, calculate the loss function based on the predicted sequence and the target sequence, update the weight parameters of the LSTM model based on the calculated loss function gradient through the backpropagation algorithm, and repeat the training process until the preset number of training times is reached or until convergence is achieved. S2. Using the ground as a reference, establish a navigation coordinate system in the VR experience space and collect real-time data of the VR headset when the human wears the VR headset, including attitude angle, acceleration, and position coordinate data in the navigation coordinate system; S3. Input the attitude angle and acceleration into the fall prediction model and extract the real-time data that triggers the fall conditions; S4. Using the sliding window method, extract the position coordinate data corresponding to the time steps within the sliding window range from the real-time data extracted in step S3 to form a sample time series, wherein the time steps within the sliding window range include the time steps of the real-time data extracted in S3; S5. After standardizing the time series of sample S4, input it into the trained LSTM neural network, and perform inverse standardization on the output to obtain the prediction result of the position coordinate data of VR head-mounted device.

2. The fall location recognition method according to claim 1, characterized in that, The standardization process specifically includes the following steps: Calculate the mean and variance of the position coordinate data for all time steps on the three coordinate axes of the navigation coordinate system. The following formula is used to convert the position coordinate data of each time step in the time series into standardized values ​​corresponding to the three coordinate axes of the navigation coordinate system: Standardized value = (Original value - Mean) / Variance; The original value is the value corresponding to the selected coordinate axis in the navigation coordinate system for the position coordinate data of each time step. The mean is the average value of the position coordinate data of all time steps corresponding to the selected coordinate axis in the navigation coordinate system. The variance is the variance of the position coordinate data of all time steps corresponding to the selected coordinate axis in the navigation coordinate system.

3. The fall location recognition method according to claim 1, characterized in that, When partitioning the dataset, a sliding window method is used, where the standardized time series within a time window is used as the input sequence, and the time series at the next time step of the input sequence is used as the target sequence.

4. The fall location recognition method according to claim 1, characterized in that, When partitioning the dataset, the sliding window method is used to divide the standardized time series within a time window into three consecutive time steps. The time series of the first two time steps are used as the input sequence, and the time series of the last time step is used as the target sequence.

5. The fall location recognition method according to claim 1, characterized in that, In step S2, the attitude angles include the pitch angle, roll angle, and yaw angle. The yaw angle represents the offset angle with respect to the Y-axis; the pitch angle represents the offset angle with respect to the X-axis; and the roll angle represents the deflection angle with respect to the Z-axis. The X-axis is the axis perpendicular to the left and right sides of the human body, the Z-axis is the axis perpendicular to the front and back sides of the human body, and the Y-axis is perpendicular to the X-axis and Z-axis. The accelerations are the accelerations AccX, AccY, and AccZ along the X, Y, and Z axes, respectively.

6. The fall location recognition method according to claim 5, characterized in that, The fall prediction model in step S3 includes an attitude angle differentiation unit and an acceleration differentiation unit. The fall prediction model performs the following steps to determine whether a fall condition is triggered: After the posture angle is input into the posture angle differentiation unit, the formula (1) can be used to distinguish the forward bending and backward falling of the human body from other postures, and obtain sqrtRP; sqrtRP calculates RPP_delta according to formula (2) and compares it with the set RPP_delta threshold, which can effectively extract the lateral and backward bending and falling movements of the human body. （1） （2） When the RPP_delta angle is greater than 30°, the acceleration is input into the acceleration differentiation unit; When the yaw angle drops to no more than 70°, the acceleration differentiation unit first calculates the combined acceleration of the Z-axis and Y-axis according to formula (3): SqrtAccYZ, based on the acceleration in the Z-axis direction and the acceleration in the Y-axis direction. Then, it calculates ACCYZZ_delta based on SqrtAccYZ, the acceleration in the Z-axis direction and formula (4), and then compares it with the normal threshold and the fall threshold to determine whether the human body has fallen. （3） （4） Observe the change in the ACCYZZ_delta angle. When ACCYZZ_delta is less than 40, it is judged as a fall, triggering the fall condition.

7. A companion robot, characterized in that, The companion robot travels to the location coordinates predicted in step S5 of claim 1 to provide protection. The companion robot carries a memory foam pad to absorb impact and energy, reducing the impact force on the head when falling.

Images