Fall detection classification method based on HMM and SVM cascade voting device

Abstract

The invention discloses a fall detection classification method based on an HMM and SVM cascade voting device, and relates to the technical field of coal mine safety, and the method comprises the steps: obtaining IMU data in a human body activity process collected by an IMU sensor; reading IMU data by using a sliding window with a fixed length, analyzing the IMU data, converting the IMU data into a floating-point number two-dimensional array, performing quaternion fusion on the floating-point number two-dimensional array to update the floating-point number two-dimensional array, and performing data filtering on the updated floating-point number two-dimensional array; step 2, outputting a falling probability based on the SMV and HMM cascade voting device, judging final output according to the set matching weight of the SVM and the HMM, if the final output is greater than or equal to a set value 0.5, determining that the output of the voting device is falling, and if the final output is smaller than the set value 0.5, circularly entering the step 1 until the output of the voting device is falling; according to the tumble detection classification method, the tumble detection accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of coal mine safety technology, and in particular to a fall detection classification method based on HMM and SVM cascade voting device. Background Art

[0002] 1. Safety issues in coal mining industry

[0003] Coal mining is a high-risk industry. Workers face a variety of potential safety hazards in underground working environments, including gas explosions, mine collapses, fires, mechanical injuries, and falls. Among them, falls are not only one of the important causes of casualties among coal miners, but may also lead to secondary accidents, resulting in more serious consequences. Therefore, timely and accurate detection and identification of falls is of great significance to protecting the lives of coal miners.

[0004] 2. Limitations of Existing Fall Detection Technologies

[0005] At present, fall detection technology is mainly divided into wearable device-based detection and video surveillance-based detection. Wearable device detection uses accelerometers, gyroscopes and other devices to monitor the worker's movement status in real time, but it faces problems such as sensor failure, improper equipment wearing and environmental interference, resulting in reduced detection accuracy. Although the detection method based on video surveillance does not rely on workers wearing equipment, in complex environments such as coal mines, factors such as light changes, dust, and occlusion have a greater impact on the quality of video images, which in turn affects the effect of fall detection. Summary of the invention

[0006] Based on the technical problems existing in the background technology, the present invention proposes a fall detection classification method based on HMM and SVM cascade voters, which improves the accuracy of fall detection.

[0007] The present invention proposes a fall detection classification method based on HMM and SVM cascade voting, comprising the following steps:

[0008] Step 1: Obtain IMU data collected by an IMU sensor during human activity, wherein the IMU data includes three-axis acceleration and three-axis angular velocity;

[0009] Step 2, feature processing module: use a fixed-length sliding window to read and parse the IMU data, convert it into a floating-point two-dimensional array, perform quaternion fusion on the floating-point two-dimensional array, generate a three-dimensional attitude angle based on roll, yaw, and pitch, make a first-order difference on the three-dimensional attitude angle, quantize the obtained first-order difference data and the three-dimensional attitude angle feature vector to update the floating-point two-dimensional array, and perform data filtering on the updated floating-point two-dimensional array;

[0010] Step 3, the first stage of the attitude detection module: SMV calculates the synthetic motion vector value based on the filtered floating point two-dimensional array. If the synthetic motion vector value is less than the set SMV violent motion threshold, the collected IMU data is cached and continues to be collected. If the synthetic motion vector value is greater than or equal to the set SMV violent motion threshold, SMV outputs the binary classification of violent motion and the timestamp. When the binary classification of violent motion is determined to be true, the second stage of the attitude detection module is entered. The SMV is a support vector machine;

[0011] Step 4, the second stage of the posture detection module: the filtered two-dimensional floating-point array is used to determine the visible and hidden state through the emission matrix and transfer matrix of the HMM. When the visible and hidden state is a fall state, the HMM outputs the probability of the second fall. At the same time, the linear SVM calculation is performed on the filtered two-dimensional floating-point array, and the SVM outputs the first fall probability. The HMM is a hidden Markov model.

[0012] Step 5: Determine the final output based on the set weight ratio of SVM and HMM. If the final output is greater than or equal to the set value 0.5, the voting machine output is a fall. If the final output is less than the set value 0.5, the loop enters step 1 until the voting machine output is a fall.

[0013] Furthermore, in step three, SMV calculates the synthetic motion vector value based on the filtered two-dimensional floating-point array, specifically: taking the most recent data in the filtered two-dimensional floating-point array, taking the three-axis acceleration for support vector machine calculation to obtain the synthetic motion vector value.

[0014] Furthermore, in step 4, the hidden Markov model is constructed as follows:

[0015] The hidden state of HMM is set to fall and non-fall, which are two features; the explicit state of HMM is set to six computable features, which are three-axis acceleration and three-axis angular velocity;

[0016] The parameters of HMM consist of three parts: initialization matrix B, transmission matrix S which is 2*6, and transfer matrix Q which is 6*6. B, S, and Q are obtained by the convergence of hidden Markov model training of the host computer.

[0017] The Viterbi path is calculated based on the B, S, and Q matrices, specifically:

[0018] Initialization: Initialize the probability of the first observation value being in each state, where the observation value is the value in the two-dimensional array of floating-point numbers after filtering of the HMM input;

[0019] Recursively: For each subsequent observation, calculate the probability of each state being the next state in the sequence of hidden states;

[0020] Termination: After processing all observations, backtrack to find the hidden state sequence with the highest probability, take the highest hidden state sequence as the probability of the second fall, and determine the second fall state as a fall state if the probability exceeds 0.5.

[0021] Furthermore, in step 5, the final output is determined by setting the weight ratio of SVM and HMM, and the final output is specifically:

[0022] Perform linear SVM calculation on the filtered two-dimensional array of floating-point numbers to obtain the first fall probability;

[0023] The HMM is assigned a weight of 0.8 and the SVM is assigned a weight of 0.2, so that the weighted sum of the second fall probability output by the HMM and the first fall probability output by the SVM is taken as the final output.

[0024] Furthermore, in step 2, the floating point two-dimensional array is specifically generated as follows:

[0025] Use fixed-length FIFO interrupts to read IMU data from the IMU's FIFO, where the FIFO interrupt threshold is set to 20 packets of data, which is a 200ms cycle at 100 Hz.

[0026] Using the sliding window method, when the set 1s window period is met, all the data in the window are parsed and converted into a floating-point two-dimensional array Sliding_data_fifo

[100] [6], that is, 100*6 matrix data.

[0027] Furthermore, in step 2, a first-order Butterworth filter is performed on the updated two-dimensional floating-point number array to obtain a filtered two-dimensional floating-point number array.

[0028] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned fall detection classification method when executing the computer program.

[0029] A computer-readable storage medium stores a plurality of classification programs, wherein the plurality of classification programs are used to be called by a processor and execute the fall detection classification method as described above.

[0030] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various media that can store program codes.

[0031] The advantage of the fall detection classification method based on HMM and SVM cascade voting device provided by the present invention is that the time series characteristics of the worker's action are captured by the hidden Markov model, and the difference between the fall action and other similar actions (such as squatting, bending over, etc.) can be more accurately identified. Combined with the classification ability of the support vector machine and the comprehensive judgment of the voting mechanism, the accuracy of fall detection can be further improved, and false alarms and missed reports can be reduced. Moreover, this embodiment comprehensively utilizes the advantages of two different models, HMM and SVM, and can maintain a high detection performance in a complex and changeable coal mine environment. Through the voting mechanism, the misjudgment of a single model in special circumstances can be effectively reduced, and the overall robustness and stability of the system can be improved. In addition, timely and accurate fall detection can trigger early warning and rescue measures in the first time, reducing the damage caused to workers by fall accidents. It helps coal mine enterprises to establish a more intelligent safety monitoring system, improve the overall safety management level, and ensure the life safety of workers. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a structural schematic diagram of the present invention;

[0033] Figure 2 This is a schematic diagram of data transmission between the terminal equipment and the upper computer system of the well monitoring platform;

[0034] Figure 3 Define the intent for the HMM visibility state; 3D attitude angles of roll, yaw, and pitch

[0035] Figures 4a to 7h It is the reference image in feature visualization processing;

[0036] Figures 4a to 4c The following is a trend graph of the posture angles Roll (roll angle), Pitch (pitch angle), and Yaw (yaw angle) changing over time in the first fall paradigm, specifically: Figure 4a This is a trend curve of the posture angle Roll (pitch angle) changing over time in the first fall paradigm. Figure 4b This is a trend curve of the Pitch (pitch angle) changing over time in the first fall paradigm. Figure 4c It is a trend curve graph of Yaw (deflection angle) changing with time in the first fall paradigm;

[0037] Figure 5a to Figure 5c The following is a trend graph of the posture angles Roll (roll angle), Pitch (pitch angle), and Yaw (yaw angle) changing over time in the second fall paradigm, specifically: Figure 5a This is a trend curve of the posture angle Roll (pitch angle) changing over time in the second fall paradigm. Figure 5bThis is a trend curve of the Pitch (pitch angle) changing over time in the second fall paradigm. Figure 5c It is a trend curve graph of Yaw (deflection angle) changing with time in the second fall paradigm;

[0038] Figures 6a to 6c The following is a trend graph of the posture angles Roll (roll angle), Pitch (pitch angle), and Yaw (yaw angle) changing over time in the third fall paradigm, specifically: Figure 6a This is a trend curve of the posture angle Roll (pitch angle) changing over time in the third fall paradigm. Figure 6b This is a trend curve of the Pitch (pitch angle) changing over time in the third fall paradigm. Figure 6c It is a trend curve diagram of Yaw (deflection angle) changing with time in the third fall paradigm;

[0039] Figures 7a to 7h It is a trend curve chart of different data features under a single set of data changing over time, specifically: Figure 7a It is a trend curve chart showing the label number changing over time, and has no actual data significance. Figure 7b It is a trend curve of SMV changing with time, that is, SMV strenuous exercise threshold; Figure 7c It is a trend curve chart showing the first-order Roll difference value in the first-order difference data changing with time; Figure 7d It is the trend curve of the second-order Roll difference value changing with time; Figure 7e It is a trend curve graph of the first-order Pitch difference value in the first-order difference data changing with time; Figure 7f It is the trend curve of the second-order Pitch difference value changing with time; Figure 7g It is a trend curve chart of the first-order Yaw difference value in the first-order difference data changing with time; Figure 7h It is a trend curve of the second-order Yaw difference value changing with time. DETAILED DESCRIPTION

[0040] Below, the technical solution of the present invention is described in detail through specific embodiments. Many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific implementation disclosed below.

[0041] In the mining process, in order to realize fall detection, the miners wear the terminal equipment around their waists. The equipment has the acquisition module, feature processing module, posture detection module and communication hardware conditions. The terminal equipment is connected to the upper computer system of the well monitoring platform through the base station (communication equipment). Therefore, the mine terminal equipment adopts the upper computer-base station (other communication intermediate layer)-terminal equipment mode. When the terminal equipment determines the fall, it is effectively reported to the ground through the data network transmission, so as to achieve the alarm effect. By using the sound and light alarm locally, the local personnel going down the mine can be reminded to rescue in the first time to upload data information. This embodiment is based on the fall detection method of HMM (hidden Markov model) and SVM (support vector machine) cascade voting device. The hidden Markov model is a statistical model used to describe time series data, which can effectively capture the time series characteristics of workers' actions. Support vector machine is a machine learning algorithm for classification with good generalization ability. By cascading HMM and SVM and combining the voting mechanism, the advantages of both can be combined to improve the accuracy and robustness of fall detection. Each module in the terminal equipment specifically implements the following scheme.

[0042] like Figures 1 to 7h As shown, the present invention proposes a fall detection classification method based on HMM and SVM cascade voting device, comprising the following steps:

[0043] Step 1, acquisition module: obtain the IMU data collected by the inertial measurement unit (IMU) sensor during human activity, wherein the IMU data includes three-axis acceleration and three-axis angular velocity;

[0044] The personnel carry out normal daily activities and simulate abnormal actions such as falling. The acquisition module of the terminal device worn by the personnel will collect the six-axis posture data (IMU data) of the personnel, with an acquisition frequency of 100HZ, and the collected data is saved in the cache. The initialization setting sampling frequency is 100hz (hertz), the unit of the accelerometer is g, the range is + / -16g, the unit of the angular velocity meter is ° / s, the range is + / -2000° / s, and the IMU is configured to high performance mode.

[0045] Step 2, feature processing module: use a fixed-length sliding window to read and parse the IMU data, convert it into a floating-point two-dimensional array, perform quaternion fusion on the floating-point two-dimensional array, generate a three-dimensional attitude angle based on roll, yaw, and pitch, make a first-order difference on the three-dimensional attitude angle, quantize the obtained first-order difference data and the three-dimensional attitude angle feature vector to update the floating-point two-dimensional array, and perform data filtering on the updated floating-point two-dimensional array;

[0046] The feature processing module is triggered by the IMU interrupt and reads the IMU data from the IMU cache. The feature processing module is divided into three stages:

[0047] In the first stage, the IMU data is read from the IMU FIFO using a fixed-length FIFO interrupt. The FIFO interrupt threshold is set to 20 packets of data, which is a 200ms cycle at 100 Hz. When the sliding window is used, all the data in the window is parsed and converted into a floating-point two-dimensional array Sliding_data_fifo

[100] [6], that is, 100*6 matrix data.

[0048] In the second stage, quaternion fusion is performed on Sliding_data_fifo

[100] [6]. Quaternion fusion can be directly implemented using existing technologies to generate three-dimensional attitude angles based on roll, yaw, and pitch. First-order differences are performed on the three-dimensional attitude angles to obtain Pitch_diff1, Yaw_diff1, and Roll_diff1. The obtained differential data and attitude angle features are vectorized, and Sliding_data_fifo

[100] [6] is updated so that its six-dimensional features are attitude angles and corresponding first-order difference values.

[0049] In the third stage, the updated Sliding_data_fifo

[100] [6] data is filtered and processed by the first-order Butterworth filter. The first-order filter has high efficiency and can solve the sensor data noise problem. The acceleration and angular velocity sensors are easily interfered by various noises during the measurement process, such as electromagnetic noise, mechanical vibration noise, etc. The stopband attenuation characteristics of the Butterworth filter are very steep, which can effectively filter out these noises higher than the required frequency range and improve the signal-to-noise ratio of the data.

[0050] This embodiment creatively uses the first-order differential features of the attitude angle as the input in the HMM, thereby improving the accuracy of the model in terms of state representation, observation time series, and state transition probability definition. The first-order differential features of the attitude angle can more keenly capture the dynamic changes of the system and improve the ability of the HMM to describe complex dynamic systems; the first-order differential features of the attitude angle have a certain robustness to small noises because these features focus on the trend of angle changes rather than instantaneous values.

[0051] Step 3, the first stage of the attitude detection module: SMV calculates the synthetic motion vector value based on the filtered floating point two-dimensional array. If the synthetic motion vector value is less than the set SMV violent motion threshold, the IMU data collected is cached and continues to be collected. If the synthetic motion vector value is greater than or equal to the set SMV violent motion threshold, SMV outputs the binary classification of violent motion and the timestamp. When the binary classification of violent motion is determined to be true, the second stage of the attitude detection module is entered. The SMV is a support vector machine;

[0052] In order to reduce the power consumption in the terminal device, this embodiment designs a cascade algorithm, in which the first level is the SMV threshold judgment, and the second level voting model judgment is performed when the first level conditions are met. This can effectively avoid unnecessary calculations and resource occupation in the terminal device. The design of the algorithm effectively reduces the power consumption of the device.

[0053] The posture detection module receives the data output from the feature processing module (i.e., the filtered two-dimensional array of floating-point numbers), and first inputs it into the first stage of the algorithm cascade to calculate the SMV (synthetic motion vector) value. If the SMV value reaches the SMV severe motion threshold Threshold_smv, it enters the second stage of the posture detection module. If it does not reach it, the intermediate data is cached and continues to be collected. The goal of the first stage of the posture detection module is to use a priori knowledge to infer that when a large SMV value does not appear, it is basically judged that no large movement has occurred, and synchronously judged that no fall has occurred.

[0054] In the first stage of the posture detection module, the input is the most recent data in the filtered Sliding_data_fifo

[100] [6]. The three-axis acceleration of the data is taken, and SMV calculation is performed to obtain the synthetic motion vector value SMVreal. The threshold Threshold_smv is used for classification. If the standard is not met, the remaining steps are skipped and the acquisition continues. The output of the first stage SMV is: 1. Flag_smv_beyond, which is a binary classification of violent motion, 2. The current timestamp corresponds to the Index index, which is used to mark the index of the current fifo, that is, the index Smv_index of the current SMV maximum value in the fifo.

[0055] Step 4, the second stage of the posture detection module: the filtered two-dimensional floating-point array is used to determine the visible and hidden state through the emission matrix and transfer matrix of the HMM. When the visible and hidden state is a fall state, the HMM outputs the probability of the second fall. At the same time, the linear SVM calculation is performed on the filtered two-dimensional floating-point array, and the SVM outputs the first fall probability. The HMM is a hidden Markov model.

[0056] When Flag_smv_beyond is determined to be true in step three, the cascade algorithm judgment begins at this time. When it is determined to be a fall according to the visible and invisible state of the HMM model, a linear SVM with a weight of 0.8 plus 0.2 is used to determine that if the value of Fall exceeds 0.5, it is determined to be a fall, and if it is less than 0.5, it is determined to be a non-fall.

[0057] The corresponding HMM model is constructed from (a1) to (a3):

[0058] (a1) The hidden state of HMM is set to fall and non-fall, which are two features; the explicit state of HMM is set to six computable features, namely three-axis acceleration and three-axis angular velocity, such as Figure 3 As shown;

[0059] (a2) The parameters of HMM consist of three parts: initialization matrix B, transmission matrix S which is 2*6, and transfer matrix Q which is 6*6. B, S, and Q are obtained by the convergence of hidden Markov model training of the host computer.

[0060] (a3) Use the Viterbi algorithm, which is a dynamic programming algorithm for finding the most likely sequence of hidden states (called the Viterbi path) in a hidden Markov model (HMM) given a sequence of observed events. The algorithm operates under the assumption that future states depend only on the current state (the Markov property).

[0061] The Viterbi path is calculated based on the B, S, and Q matrices, specifically (a3-1) to (a3-3):

[0062] (a3-1) Initialization: Initialize the probability of the first observation value being in each state, where the observation value is the value in the two-dimensional array of floating-point numbers after filtering of the HMM input;

[0063] (a3-2) Recursion: For each subsequent observation value, calculate the probability of each state becoming the next state in the hidden state sequence. The calculation process uses the HMM model's own calculation method, but the matrices and features involved are specifically set for this embodiment. The matrices are specifically: initialization matrix B, emission matrix S is a 2*6 matrix, and transfer matrix Q is a 6*6 matrix. The features are specifically: the explicit state of the HMM is set to six computable features, namely three-axis acceleration and three-axis angular velocity; this step involves checking the probability of converting from each previous state to the current state and selecting the highest probability;

[0064] (a3-3) Termination: After processing all observations, backtrack to find the hidden state sequence with the highest probability, take the highest hidden state sequence as the probability of the second fall, and determine the second fall state as a fall state if the probability exceeds 0.5.

[0065] Through (a3-1) to (a3-3), the recursive process of this embodiment effectively decomposes the problem into simpler, overlapping sub-problems, and stores their solutions to avoid redundant calculations. In (a1) to (a3), the iteration of the dimensional bit algorithm based on the actual IMU data is adopted, where the data type is still consistent with the floating point two-dimensional array Sliding_data_fifo

[100] [6].

[0066] After HMM outputs the probability of the second fall, the linear SVM calculation is performed on the filtered two-dimensional array of floating-point numbers to obtain the probability of the first fall; the linear one-dimensional operator T in the SVM also uses the HMM algorithm, but the difference between the HMM algorithm here and the above HMM algorithm is that the explicit state parameters used are different. The explicit state parameters used by the HMM algorithm here are: Pitch_diff1_cnt, corresponding to the number of jumps under Std_time_windows; Pitch_std, corresponding to the standard deviation of the pitch Ptich under Std_time_windows; Roll_diff1_std, corresponding to the standard deviation of the first-order difference of the roll Roll parameter under Std_time_windows; Std_time_windows, the time window for calculating the standard deviation, currently based on the continuous state after the fall, 1.5s, 100hz, 150 points, that is, 150 time steps. (This parameter can be used as an expansion window to adjust the sensitivity).

[0067] Step 5: Determine the final output based on the set weight ratio of SVM and HMM. If the final output is greater than or equal to the set value 0.5, the voting machine output is a fall. If the final output is less than the set value 0.5, the voting machine output is a non-fall.

[0068] The HMM is assigned a weight of 0.8 and the SVM is assigned a weight of 0.2, so that the weighted sum of the second fall probability output by the HMM and the first fall probability output by the SVM is used as the final output. If the final output is greater than or equal to the set value of 0.5, the voting machine outputs a fall, which is judged to be an abnormal posture. The abnormal posture terminal emits an audible and visual alarm and sends the detection result to the base station. The abnormal posture is transmitted to the well monitoring platform through the base station, and the well staff performs the next step of processing.

[0069] According to steps one to five, this embodiment captures the time series characteristics of the worker's actions through the hidden Markov model, and can more accurately identify the difference between the fall action and other similar actions (such as squatting, bending over, etc.). Combined with the classification ability of the support vector machine and the comprehensive judgment of the voting mechanism, the accuracy of fall detection can be further improved, and false alarms and missed reports can be reduced. Moreover, this embodiment comprehensively utilizes the advantages of two different models, HMM and SVM, and can maintain a high detection performance in a complex and changeable coal mine environment. Through the voting mechanism, the misjudgment of a single model in special circumstances can be effectively reduced, and the overall robustness and stability of the system can be improved. In addition, timely and accurate fall detection can trigger early warning and rescue measures in the first time, reducing the damage caused to workers by fall accidents. It helps coal mine enterprises to establish a more intelligent safety monitoring system, improve the overall safety management level, and ensure the life safety of workers.

[0070] As an example:

[0071] Step 1: The person wears the terminal device correctly around the waist.

[0072] Step 2: The personnel carry out normal daily activities and simulate abnormal actions such as falling. The acquisition module of the terminal device will collect the six-axis posture data of the personnel at a frequency of 100HZ, and the collected data will be stored in the cache.

[0073] Step 3: The feature processing module is triggered by the IMU interrupt and reads the IMU data from the cache. The IMU data is converted into attitude angles and first-order difference data, a total of six-dimensional features. After passing through the Badyworth low-order filter, it is used as the input for the next step.

[0074] Step 4: First, determine whether the two-dimensional array of floating-point numbers after filtering meets the SMV severe motion threshold. If it does, enter the next stage. At this time, the binary classification Flag_smv_beyond for severe motion is True. The Viterbi path is calculated through the given B, S, and Q matrices, and the hidden state under the HMM algorithm is recursively obtained, that is, the probability of the second fall. If the probability exceeds 0.5, it is determined to be a fall. The training samples used by the B, S, and Q matrices are the action recordings under the demonstration format shown in Table 1.

[0075] Secondly, a linear SVM calculation is performed on the feature vectors in the filtered two-dimensional array of floating-point numbers, where the linear one-dimensional operator T in the SVM is also the fitted SVM model trained by Table 1. The fall probability can also be obtained. The HMM algorithm is assigned a weight of 0.8 and the SVM is assigned a weight of 0.2 to obtain the final output. If the final output exceeds 0.5, it is judged as a fall, and if it is lower than 0.5, no processing is performed.

[0076] Step 5: In the above training samples, the data set used for posture detection module training is collected independently. The action paradigm of the collected data is shown in Table 1. The developer wears a terminal on the waist to simulate the corresponding actions according to the action paradigm. There are 20 action paradigms, which are divided into daily activity action paradigms and fall action paradigms. The samples collected 15 people of different body shapes, and each action paradigm was simulated 30 times, totaling 9,000 sets of data.

[0077] Table 1 Human body posture action paradigm

[0078]

[0079]

[0080] Step 6: When the posture is determined to be abnormal, the terminal will send the abnormal status to the base station (or other communication intermediate equipment) through the communication module, and the base station will transmit it to the well monitoring platform.

[0081] This embodiment uses the hidden Markov model's sensitivity to binary states and the segmented cascade method to reduce power consumption, and completes the fall detection of terminal devices worn in mines based on the voting method of the SVM linear machine learning algorithm, and improves the detection accuracy and system robustness by integrating the advantages of different algorithms. In addition, this embodiment has strong data filtering capabilities, strong adaptability, and strong robustness, and is particularly suitable for complex environments such as underground coal mines. It can be stably executed under harsh conditions, providing reliable protection for the safe production of miners.

[0082] like Figures 4a to 6c , collect the trends of the attitude angles Roll (roll angle), Pitch (tilt angle), and Yaw (yaw angle) over time in the three fall paradigms ( Figures 4a to 4c For a group, Figure 5a to Figure 5c For a group, Figures 6a to 6c as a group), Figures 4a to 6c The horizontal axis is the time step, and the single time step is 1 / 100s. The vertical axis indicates different units according to different angles. Figures 4a to 6c It can be seen from these three groups that: 1. The Roll angle drops sharply at the moment of falling, 2. The Pitch fluctuates periodically before falling, drops instantly during falling, and changes relatively stably afterwards, 3. The change of Yaw is similar to that of Pitch.

[0083] like Figures 7a to 7h These are different data features for a single set of data. The horizontal axis is also the time step, with the unit being 1 / 100s, and the vertical axis is the corresponding scale unit.

[0084] from Figures 4a to 7h The following conclusions can be drawn: By fusing the posture angles, more obvious fall features can be obtained under visualization. Different differences can improve the stability of the data, filter the time trend items, and better extract features.

[0085] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A fall detection classification method based on HMM and SVM cascade voting device, comprising the following steps: Step 1: Obtain IMU data collected by an IMU sensor during human activity, wherein the IMU data includes three-axis acceleration and three-axis angular velocity; Step 2, feature processing module: use a fixed-length sliding window to read and parse the IMU data, convert it into a floating-point two-dimensional array, perform quaternion fusion on the floating-point two-dimensional array, generate a three-dimensional attitude angle based on roll, yaw, and pitch, make a first-order difference on the three-dimensional attitude angle, quantize the obtained first-order difference data and the three-dimensional attitude angle feature vector to update the floating-point two-dimensional array, and perform data filtering on the updated floating-point two-dimensional array; Step 3, the first stage of the attitude detection module: SMV calculates the synthetic motion vector value based on the filtered floating point two-dimensional array. If the synthetic motion vector value is less than the set SMV violent motion threshold, the collected IMU data is cached and continues to be collected. If the synthetic motion vector value is greater than or equal to the set SMV violent motion threshold, SMV outputs the binary classification of violent motion and the timestamp. When the binary classification of violent motion is determined to be true, the second stage of the attitude detection module is entered. The SMV is a support vector machine; Step 4, the second stage of the posture detection module: the filtered two-dimensional floating-point array is used to determine the visible and hidden state through the emission matrix and transfer matrix of the HMM. When the visible and hidden state is a fall state, the HMM outputs the probability of the second fall. At the same time, the linear SVM calculation is performed on the filtered two-dimensional floating-point array, and the SVM outputs the first fall probability. The HMM is a hidden Markov model. Step 5: Determine the final output based on the set weight ratio of SVM and HMM. If the final output is greater than or equal to the set value 0.5, the voting machine output is a fall. If the final output is less than the set value 0.5, the loop enters step 1 until the voting machine output is a fall.

2. The fall detection classification method based on HMM and SVM cascade voting device according to claim 1 is characterized in that: In step three, SMV calculates the synthetic motion vector value based on the filtered two-dimensional floating-point array, specifically: using the most recent data in the filtered two-dimensional floating-point array, taking the three-axis acceleration for support vector machine calculation to obtain the synthetic motion vector value.

3. The fall detection classification method based on HMM and SVM cascade voting device according to claim 1 is characterized in that: In step 4, the hidden Markov model is constructed as follows: The hidden states of the HMM are set to fall and non-fall, which are 2 features; The explicit states of the HMM are set to six computable features, namely, three-axis acceleration and three-axis angular velocity; The parameters of HMM consist of three parts: initialization matrix B, transmission matrix S which is 2*6, and transfer matrix Q which is 6*6. B, S, and Q are obtained by the convergence of hidden Markov model training of the host computer. The Viterbi path is calculated based on the B, S, and Q matrices, specifically: Initialization: Initialize the probability of the first observation value being in each state, where the observation value is the value in the two-dimensional array of floating-point numbers after filtering of the HMM input; Recursively: For each subsequent observation, calculate the probability of each state being the next state in the sequence of hidden states; Termination: After processing all observations, backtrack to find the hidden state sequence with the highest probability, take the highest hidden state sequence as the probability of the second fall, and determine the second fall state as a fall state if the probability exceeds 0.

5.

4. The fall detection classification method based on HMM and SVM cascade voting device according to claim 3 is characterized in that: In step 5, the final output is determined by the set ratio weights of SVM and HMM. The final output is specifically: Perform linear SVM calculation on the filtered two-dimensional array of floating-point numbers to obtain the first fall probability; The HMM is assigned a weight of 0.8 and the SVM is assigned a weight of 0.2, so that the weighted sum of the second fall probability output by the HMM and the first fall probability output by the SVM is taken as the final output.

5. The fall detection classification method based on HMM and SVM cascade voting device according to claim 1 is characterized in that: In step 2, the floating point two-dimensional array is generated as follows: Use fixed-length FIFO interrupts to read IMU data from the IMU's FIFO, where the FIFO interrupt threshold is set to 20 packets of data, which is a 200ms cycle at 100 Hz. Using the sliding window method, when the set 1s window period is met, all the data in the window are parsed and converted into a floating-point two-dimensional array Sliding_data_fifo[100][6], that is, 100*6 matrix data.

6. The fall detection classification method based on HMM and SVM cascade voting device according to claim 1 is characterized in that: In step 2, a first-order Butterworth filter is performed on the updated two-dimensional floating-point number array to obtain a filtered two-dimensional floating-point number array.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the fall detection classification method according to any one of claims 1 to 6 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of classification programs, and the plurality of classification programs are used to be called by a processor and execute the fall detection classification method according to any one of claims 1 to 6.

Images