A Human Fall Detection Method and System Based on an Improved Whale Algorithm

Abstract

This disclosure provides a method and system for human fall detection based on an improved whale algorithm, relating to the field of fall detection technology. The method includes acquiring human behavior data, which consists of triaxial acceleration, angular velocity, and angle values; extracting time-domain features from the human behavior data, converting the extracted time-domain features into frequency-domain features, dividing the frequency-domain features into high-frequency and low-frequency signals, and using a three-layer wavelet packet transform to mine hidden features in the high-frequency and low-frequency signals, extracting the energy spectrum of the third-layer nodes of the wavelet packet as time-frequency domain features; using an improved whale optimization algorithm to remove redundant features from the initial time-domain, frequency-domain, and time-frequency-domain features, performing feature selection, inputting the selected features into an SVM model, and outputting the recognition and detection result of whether a fall has occurred; this disclosure improves the recognition accuracy.

Description

Technical Field

[0001] This disclosure relates to the field of fall detection technology, specifically to a human fall detection method and system based on an improved whale algorithm. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] With the increasing aging of society and the prevalent "4-2-1" family model, the number of elderly people will continue to grow. Conversely, existing public resources for elderly care are insufficient to support the future care needs of the aging population; therefore, family-based elder care will be the primary mode of care in the future. In a family setting, monitoring the health status of the elderly is particularly important.

[0004] Statistics show that falls have become the leading cause of health risks for the elderly. Currently, fall detection systems can be broadly categorized into three types based on the data source: user-triggered, computer vision-based, and wearable device-based fall detection systems. Wearable device-based fall detection systems primarily consist of one or more posture sensors worn on specific parts of the user's body to collect motion data. Compared to the first two types of systems, wearable devices do not restrict the user's range of motion, and most posture sensors are inexpensive. Therefore, research on wearable fall detection systems has received increasing attention.

[0005] Research on wearable fall detection systems still faces several challenges, such as finding features that can effectively distinguish between fall behavior and everyday behavior. Because the same behavior varies from individual to individual, the collected data cannot cover all states, potentially leading to the loss of some effective features. Furthermore, human behavior is continuous, with no clear boundary between actions, further complicating fall detection and recognition. These issues render traditional feature construction methods ineffective in identifying fall behavior, and the limited variety of feature samples prevents the acquisition of more relevant and effective features. Traditional feature selection methods struggle to obtain high-quality feature combinations. The whale optimization algorithm, inspired by humpback whale predation behavior, is an intelligent optimization algorithm with advantages of efficiency and ease of implementation, widely used to improve feature selection performance. However, due to the single leading individual, weak exploration ability in later iterations, and dispersed individual positions in early iterations, the original whale algorithm suffers from insufficient global exploration ability and is prone to getting trapped in local optima, especially when dealing with high-dimensional or complex real-world applications. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure proposes a method and system for human fall detection based on an improved whale algorithm. The improved whale algorithm obtains more effective features that are highly relevant to the target, thereby achieving high-precision recognition of fall behavior.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions:

[0008] A human fall detection method based on an improved whale algorithm includes:

[0009] Acquire human behavioral data, which consists of triaxial acceleration, angular velocity, and angle values;

[0010] The temporal features of human behavior data are extracted, the extracted temporal features are converted into frequency features, the frequency features are divided into high-frequency signals and low-frequency signals, and the hidden features in the high-frequency signals and low-frequency signals are mined using three-layer wavelet packet transform. The energy spectrum of the third layer node of the wavelet packet is extracted as the temporal and frequency domain features.

[0011] The improved whale optimization algorithm is used to remove redundant features from the initial time domain, frequency domain, and time-frequency domain features. Feature selection is then performed, and the selected features are input into the SVM model to output the recognition and detection results of whether a fall has occurred.

[0012] According to some embodiments, the present disclosure adopts the following technical solutions:

[0013] A human fall detection system based on an improved whale algorithm includes:

[0014] The data acquisition module is used to acquire human behavior data, which consists of triaxial acceleration, angular velocity, and angle values.

[0015] The feature selection module is used to extract the time-domain features of human behavior data, convert the extracted time-domain features into frequency-domain features, divide the frequency-domain features into high-frequency signals and low-frequency signals, and use three-layer wavelet packet transform to mine the hidden features in the high-frequency signals and low-frequency signals, and extract the energy spectrum of the third layer node of the wavelet packet as the time-frequency domain features.

[0016] The improved whale optimization algorithm is used to remove redundant features from the initial time domain, frequency domain, and time-frequency domain features for feature selection.

[0017] The detection module is used to input the selected features into the SVM model and output the detection result of whether a fall has occurred.

[0018] According to some embodiments, the present disclosure adopts the following technical solutions:

[0019] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the human fall detection method based on the improved whale algorithm.

[0020] According to some embodiments, the present disclosure adopts the following technical solutions:

[0021] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the human fall detection method based on the improved whale algorithm.

[0022] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0023] This disclosure provides a method and system for human fall detection based on an improved whale algorithm, named MSWOA. An Adaptive Multipopulation Merging Strategy (AMS) is proposed to enhance the global search capability of the population, enabling the discovery of feature subsets with multiple combinations. A Double Spiral Updating Strategy (DSS) is also proposed to improve the randomness of individual populations, allowing for continuous searching and updating even in later iterations. A Baleen Neighborhood Exploitation Strategy (BES) improves the overall convergence speed of the algorithm. The introduction of these three strategies enables the MSWOA-based feature selection method to obtain feature combinations with higher recognition accuracy. Attached Figure Description

[0024] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0025] Figure 1 The flowchart is for the MSWOA algorithm;

[0026] Figure 2 This is a graph showing the spiral trajectory of the particle.

[0027] Figure 3 A comparison of convergence curves for WOA variant algorithms;

[0028] in, Figure 3 (a) shows the convergence curves of the seven algorithms on the 9_Tumors dataset;

[0029] Figure 3 (b) shows the convergence curves of the seven algorithms on the 11_Tumors dataset;

[0030] Figure 3 (c) shows the convergence curves of the seven algorithms on the Isolet dataset;

[0031] Figure 3 (d) in the figure represents the convergence curves of the seven algorithms on the Lung_discrete dataset;

[0032] Figure 3 (e) in the figure represents the convergence curves of the seven algorithms on the Lung_Cancer dataset;

[0033] Figure 3 (f) in the figure represents the convergence curves of the seven algorithms on the Semion dataset;

[0034] Figure 4 This is a circuit simulation connection diagram for hardware devices;

[0035] Figure 5 Here is a flowchart of the fall detection system;

[0036] Figure 6 Experimental process and environmental scene diagrams for some behaviors;

[0037] Figure 7 A graph comparing the accuracy of fall behavior detection results. Detailed Implementation

[0038] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0040] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0041] Example 1

[0042] One embodiment of this disclosure provides a method for detecting human falls based on an improved whale algorithm, the steps of which include:

[0043] Step 1: Acquire human behavioral data, which consists of triaxial acceleration, angular velocity, and angle values;

[0044] Step 2: Extract the time-domain features of human behavior data, convert the extracted time-domain features into frequency-domain features, divide the frequency-domain features into high-frequency signals and low-frequency signals, and use three-layer wavelet packet transform to mine the hidden features in the high-frequency signals and low-frequency signals. Extract the energy spectrum of the third layer node of the wavelet packet as the time-frequency domain features.

[0045] Step 3: Use the improved whale optimization algorithm to remove redundant features from the initial time domain, frequency domain, and time-frequency domain features, perform feature selection, input the selected features into the SVM model, and output the recognition and detection results of whether a fall has occurred.

[0046] As one embodiment, the training steps of a human fall detection method based on an improved whale algorithm include:

[0047] Step 1: Collect behavioral data and construct the original posture dataset: First, the detection module is worn on the waist of the experimenter, the sampling frequency is set to 20Hz, and the single sampling time is set to 10s. Data is collected sequentially according to the designed 13 daily life behaviors and 13 fall behaviors. The behavior types are shown in Table 1 and Table 2. Then, the obtained data consists of triaxial acceleration, angular velocity and angle values, that is, each sample has 200 sampling points and 9 original features. Finally, the data samples obtained under different behaviors are summarized and organized to form the original data sample set.

[0048] Human behavior data includes normal daily life behaviors and abnormal fall behaviors, as shown in Tables 1 and 2, which are the data collection types for normal daily life behaviors and abnormal fall behaviors, respectively.

[0049] Table 1 Data Types of Normal Daily Life Behaviors

[0050] Serial Number Daily life behaviors Number of experiments Collection time 1 Walk slowly 10 10s 2 Hurry up 10 10s 3 jogging 10 10s 4 Slowly go up and down the stairs 10 10s 5 Go up and down stairs quickly 10 10s 6 Slowly sit down and stand up. 10 10s 7 Sit down and stand up quickly 10 10s 8 When trying to get up from a chair, he suddenly slumped down. 10 10s 9 Sit down slowly, then lie down again and sit up. 10 10s 10 Stand up – slowly bend your knees – stand up 10 10s 11 Standing – Bending over (without bending knees) – Standing 10 10s 12 Tripped a little (but didn't fall). 10 10s 13 Take a light jump 10 10s

[0051] Table 2 Data Types of Abnormal Falling Behaviors

[0052]

[0053]

[0054] Step 2: Feature extraction of the collected raw data samples. Thirteen time-domain features were extracted using statistical methods, namely mean, standard deviation, skewness, maximum value, minimum value, peak-to-peak value, average variation, root mean square, kurtosis, waveform factor, peak factor, impulse factor, and margin factor. The feature expressions are shown in Table 3.

[0055] Table 3 Expressions of Time-Domain Feature Parameters

[0056]

[0057] In Table 3, x(n), n=1,2,...,N is a given discrete time series, and N represents the number of sample points of the signal x(n).

[0058] Step 3: Use the Fast Fourier Transform (FFT) method to convert the time-domain signal into a frequency-domain signal to extract richer attitude features. Attitude features refer to the features extracted from attitude behavior; "richer" refers to features extracted from different types and angles, and "richer features" ensures that features that effectively distinguish attitude behavior are not missed. The calculation formula for the Fast Fourier Transform is as follows:

[0059]

[0060]

[0061] In the formula, x(n), n=1,2,...,N is a given discrete-time series, and N represents the number of sample points of the signal x(n). This represents the butterfly factor. The mean, variance, maximum, minimum, peak-to-peak value, and average variation of the energy amplitude are extracted from the frequency domain signal as the second part of the features.

[0062] Step 4: Use a three-layer wavelet packet transform to perform in-depth analysis of high-frequency and low-frequency signals, uncovering hidden features in the data. The wavelet packet transform decomposes the original time-domain signal into high-frequency and low-frequency components. The features it obtains differ from the surface features such as maximum and average acceleration extracted through time-frequency analysis. These features cannot be obtained through direct observation and require relevant analysis for extraction; hence, they are called hidden features. The energy spectrum of the third-layer nodes of the wavelet packet is extracted as a time-frequency domain feature. The energy calculation formula for the nodes after wavelet packet decomposition is as follows:

[0063]

[0064] In the formula, M represents the length of the i-th frequency band. Let E represent the coefficient of the nth wavelet packet corresponding to node (i,k), E represent the energy value of the corresponding node, and i represent the i-th layer.

[0065] By integrating the time-frequency domain features with the time-domain and frequency-domain features, a 243-dimensional initial pose dataset is constructed.

[0066] Step 5: Use a feature selection method based on the improved whale optimization algorithm to remove redundant features from the initial dataset. The flowchart of the improved algorithm MSWOA is shown below. Figure 1 As shown.

[0067] Specifically:

[0068] 1) Set the initial parameters of the whale population, such as the maximum number of iterations MaxIter and the number of particles N.

[0069] The population is initialized using the following formula:

[0070] X = X min +(X max -X min )*rand (4)

[0071] Where X represents the current individual in the population, and rand represents a random number between [0,1].

[0072] 2) At the start of the iteration, calculate the fitness value of each individual and update the globally optimal individual X. * .

[0073] The population is divided into multiple subpopulations according to the Adaptive Multipopulation Merging Strategy (AMS). First, the number of subpopulations is determined. The number of subpopulations gradually changes according to the iteration cycle to meet the needs of global and local search at different stages. The calculation formula is as follows:

[0074]

[0075] In the formula, K is the number of subpopulations, floor is the floor function, N is the population size, t is the current iteration number, and MaxIter is the maximum number of iterations. γ is a parameter in [0,1], set according to the population size.

[0076] 3) Then, determine the individual composition of each subpopulation and update the optimal individual X of the subpopulation. m * To balance global and local search, a new subpopulation individual selection mechanism was developed, taking into account both the distance between individuals and their fitness values. Based on the normalized individual fitness values ​​and the distance between each individual and the global best individual, the comprehensive value of each individual was calculated, as shown in formula (6). The results were then sorted from highest to lowest and divided into different subpopulations. Within each subpopulation, the individual fitness values ​​were compared, and the optimal individual X of the subpopulation was selected. m * .

[0077] FD i =(1-fit) i )+dis i,i=1, 2...,N (6)

[0078] In the formula, fit is the normalized fitness value, and dis is the distance between the normalized individual and the global best individual.

[0079] In addition, a random reassignment mechanism was introduced to facilitate information exchange between subpopulations. Every R iterations, individuals are no longer selected using the aforementioned mechanism, but are randomly assigned to different subpopulations. This step is shown in Table 4.

[0080] Table 4 Random Selection Steps

[0081]

[0082] 4) Individual positions are updated sequentially within each subpopulation. The method of updating individual positions is divided into three types based on the values ​​of parameters A and p. When p < 0.5 and |A| < 1, the position is updated according to formula (7); when p > 0.5, the position is updated according to formula (8).

[0083]

[0084]

[0085] When p < 0.5 and |A| > 1, to increase the random search capability of individuals and avoid the problem of search stagnation in the later stages of iteration, the characteristics of the golden spiral curve—which infinitely contracts inward and infinitely diverges outward—are utilized. Individuals start from their current position and randomly spiral outward to update to nearby or distant positions, unaffected by the number of iterations or the relative positions among the population. This is called the Double Spiral Updating Strategy (DSS), and the update formula is as follows:

[0086] X m (t+1)=r·e θ cos(2πθ)+X m (t) (9)

[0087] r = |W·X rand -X m (t)| (10)

[0088] In the formula, X m Let X represent the current individual in the m-th subpopulation. rand Let θ represent a random individual in another subpopulation, θ control the radian of the spiral, and W is a random vector in [0,1].

[0089] Under the combined effect of the AMS and DSS strategies, the search path of each subpopulation exhibits an inward spiral trend, while the exploration direction of individual individuals exhibits an outward spiral trend. The combination of these two strategies creates an interacting spiral curve, enabling the discovery of more potential candidate solutions in the current space, such as... Figure 2 .

[0090] 5) Explore further in the vicinity of the current best individual in the subpopulation. If a better solution is found, update the best individual in the subpopulation. Inspired by the baleen-assisted feeding habits of humpback whales, the Baleen Neighborhood Exploitation Strategy (BES) is proposed. A mathematical model is constructed through quadratic interpolation to help the best individual in the subpopulation explore a better position in the neighborhood. Especially in the early optimization stage, this is beneficial to enhance individual quality and accelerate convergence. First, calculate the new individual position X according to formula (11). q .

[0091]

[0092] In the formula, q j Represents X q (q1, q2, ..., q D The j-th dimension of X m * (g1, g2..., g D ) and X m (m1, m2..., m D X and X' are the best individual and the current individual in the m-th subpopulation, respectively. r (r1, r2..., r D ) is a random individual, F(X) m * ), F(X) m ) and F(X r ) are X m * X m and X r The fitness value.

[0093] Then, for the best individual X in the current subpopulation m * and X q Perform crossover operations to obtain intermediate solution X c The formula is as follows:

[0094]

[0095] In the formula, c j Represents X c (c1, c2, ..., c D The j-th dimension of ).

[0096] Finally, calculate and compare X. c and X m * If the updated fitness value is better than the current best fitness value, then the best individual in the subpopulation is updated to X. c .

[0097] Finally, determine whether all K subpopulations have completed position updates. If they have, check the iteration count; otherwise, return to step 4.

[0098] Determine if the maximum iteration period has been reached. If it has, end the iteration and obtain the filtered feature subset; otherwise, return to step 2.

[0099] The obtained feature subset is divided into a training set and a test set in an 8:2 ratio. The training set is input into the classification model SVM for model training. After training, the test set is input into the SVM for fall detection model testing.

[0100] This disclosure utilizes a designed fall detection module to collect data and create a posture dataset. The data collection experiment involved ten participants, including three women and seven men. Due to the inherent risks of falling, elderly volunteers could not be recruited; therefore, all participants were students with an average age of 22, an average height of 174.7 cm, and an average weight of 65.2 kg. The fall detection system's sampling frequency was set to 20 Hz, and the single sampling time was set to 10 seconds. The collected raw data underwent preprocessing, and features were extracted using time-domain, frequency-domain, and time-frequency-domain analysis methods to obtain a 243-dimensional posture dataset.

[0101] The improved whale algorithm in the experiment was named MSWOA (Multispiral Whale Optimization Algorithm). To verify the performance of MSWOA feature selection, a comparative experiment was conducted on the UCI benchmark dataset with six WOA whale variant algorithms. Figure 3The convergence curves of seven algorithms are shown. Based on the experimental results, MSWOA exhibits a faster convergence speed and achieves the best fitness value. On the 11_Tumors, Isolet, and Lung_discrete datasets, MSWOA consistently outperforms other algorithms and also boasts a better fitness value than the other comparative methods. This is attributed to AMS (Adaptive Multipopulation Merging Strategy), which expands the search scope in the early stages, identifying more promising candidate solutions and search directions. On the 9_Tumors, Orlraws10P, and Semion datasets, WOA (Whale Optimization Algorithm), RDWOA (RD Whale Optimization Algorithm), QWOA (Quantum Whale Optimization Algorithm), and SBWOA (Spatial Bound Whale Optimization Algorithm) perform slightly better than MSWOA in the early iterations. However, driven by the BES (Bone Neighborhood Deep Development) strategy, MSWOA quickly discovers a better solution and surpasses MSWOA. Besides its convergence speed, MSWOA's ability to escape local optima and avoid search stagnation is also outstanding after introducing the DSS (Bidirectional Spiral Update) strategy. On the Colon, Lung_Cancer, and Yale datasets, MSWOA can still find new search directions and continuously discover better solutions after a brief stagnation, while other algorithms get stuck in local optima, stop updating, and even those with the ability to escape local optima still perform worse than MSWOA.

[0102] In summary, the AMS strategy improves the global search capability of the population, enabling the discovery of feature subsets with multiple combinations; the DSS strategy increases the randomness of individual populations, allowing for continuous searching and updating even in later iterations; and the BES strategy improves the overall convergence speed of the algorithm. The introduction of these three strategies allows the MSWOA-based feature selection method to obtain feature combinations with higher recognition accuracy.

[0103] The actual circuitry and appearance of the fall detection system are as follows: Figure 4 As shown, the overall process is as follows: Figure 5 After the system board is powered on, the main control board and attitude sensors enter a cyclic working state. The attitude sensors collect behavioral data and send it to the system main controller for data preprocessing to obtain standardized data packets. The data acquisition process is as follows: Figure 6As shown, the experimenter wore the detection device on his waist. The device and the computer were connected via a wireless serial port. After starting the operation, the data was directly transmitted to the computer and converted into decimal data using a decoding formula. Each sample was then saved as a separate xls file after decoding.

[0104] After obtaining the raw data files, feature extraction is first used to mine rich pose features from the data, forming a preliminary pose dataset. Then, the MSWOA-optimized feature selection method is used to accurately remove redundant, repetitive, and invalid features from a large number of features, obtaining the subset of features with the highest relevance to the behavior, i.e., the processed pose dataset. Finally, the pose dataset is divided into a training set and a test set. The training set is input into the SVM model for training. After training, the test set is input to perform behavior judgment and output the recognition results.

[0105] The pose dataset processed by MSWOA-FS was input into the classification model, and compared with the original dataset and the pose dataset selected by WOA features. The experimental results are shown in Table 5. The table shows that the fall detection system using the MSWOA-FS method achieves a recognition accuracy of 95.60%, and the F-score also reaches 95.66%, which is 3 percentage points higher than the WOA-FS algorithm (a feature selection method based on the whale algorithm). Analysis of the confusion matrix results shows that MSWOA-FS correctly predicted the most fall samples, had the lowest average number of misclassified fall samples, and the lowest overall number of mispredictions (FN+FP), indicating that the algorithm has strong recognition capabilities for both fall samples and everyday behavior samples. Therefore, the fall detection system using the MSWOA algorithm for feature selection can reliably distinguish between fall and non-fall behaviors.

[0106] Table 5 shows the recognition results on the pose dataset.

[0107]

[0108] Example 2

[0109] One embodiment of this disclosure provides a human fall detection system based on an improved whale algorithm, comprising:

[0110] The data acquisition module is used to acquire human behavior data, which consists of triaxial acceleration, angular velocity, and angle values.

[0111] The feature selection module is used to extract the time-domain features of human behavior data, convert the extracted time-domain features into frequency-domain features, divide the frequency-domain features into high-frequency signals and low-frequency signals, and use three-layer wavelet packet transform to mine the hidden features in the high-frequency signals and low-frequency signals, and extract the energy spectrum of the third layer node of the wavelet packet as the time-frequency domain features.

[0112] The improved whale optimization algorithm is used to remove redundant features from the initial time domain, frequency domain, and time-frequency domain features for feature selection.

[0113] The detection module is used to input the selected features into the SVM model and output the detection result of whether a fall has occurred.

[0114] Example 3

[0115] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the human fall detection method based on the improved whale algorithm.

[0116] Example 4

[0117] One embodiment of this disclosure provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the human fall detection method based on the improved whale algorithm.

[0118] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0120] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A human fall detection method based on whale optimization algorithm, characterized in that, The method comprises the following steps: acquiring human behavior data, wherein the behavior data is composed of three-axis acceleration, angular velocity and angle value; extracting time domain features of the human behavior data, converting the extracted time domain features into frequency domain features, dividing the frequency domain features into high frequency signals and low frequency signals, and using three-layer wavelet packet transform to mine the hidden features in the high frequency signals and the low frequency signals, and extracting the third layer node energy spectrum of the wavelet packet as time-frequency domain features; using an improved whale optimization algorithm to remove redundant features in the initial time domain, frequency domain and time-frequency domain features, performing feature selection, and inputting the features after feature selection into an SVM model to output a recognition detection result of whether a fall behavior occurs; wherein the method of using the improved whale optimization algorithm to remove redundant features in the initial time domain, frequency domain and time-frequency domain features and performing feature selection comprises: (1) setting initial parameters of a whale population, performing iteration, calculating fitness values of each individual, and updating a global optimal individual; (2) dividing the population into multiple sub-populations according to an adaptive multi-population merging strategy, determining the number of sub-populations, determining the individual composition of each sub-population, updating the optimal individual of each sub-population, formulating a new sub-population individual selection mechanism by considering the distance and fitness value between individuals, calculating the comprehensive value of each individual according to the normalized individual fitness value and the distance between the individual and the global optimal individual; (3) introducing a random recombination mechanism to promote information exchange between sub-populations, and individuals are randomly assigned to different sub-populations; (4) each sub-population sequentially updates the position of individuals; according to an iteration completion condition, the update is completed, and finally the features are obtained.

2. The human fall detection method based on whale optimization algorithm according to claim 1, wherein, The human behavior data includes daily life normal behavior and fall abnormal behavior data.

3. The whale optimization algorithm-based human fall detection method of claim 1, wherein, The time domain features include mean value, standard deviation, skewness, maximum value, minimum value, peak-peak value, average change amount, root mean square, kurtosis, waveform factor, peak factor, pulse factor and margin factor.

4. The human fall detection method based on whale optimization algorithm according to claim 1, wherein, The time domain features are converted into frequency domain features using a fast Fourier transform method, and the mean value, variance, maximum value, minimum value, peak-peak value and average change amount of the energy amplitude are extracted from the frequency domain signals as the second part of the features. 5.The human fall detection method based on whale optimization algorithm of claim 1, wherein, The three-layer wavelet packet transform is used to mine the high frequency signals and the low frequency signals, mine the hidden features in the data, and extract the third layer node energy spectrum of the wavelet packet as the time-frequency domain features. In the formula, M Indicates the first i Each frequency band length, Represents the nodes after wavelet packet decomposition ( i , k The corresponding number n One coefficient, E This represents the energy value of the corresponding node. i Indicates the first i layer.

6. The human fall detection method based on whale optimization algorithm according to claim 1, wherein, The population is divided into multiple sub-populations according to the adaptive multi-population merging strategy, and the number of sub-populations is determined in the following manner: wherein K is the number of subpopulations, floor is the floor function, N is the population size, t is the current iteration number, MaxIter is the maximum iteration number; γ is a parameter in [0, 1] that is set according to the population size.

7. A human fall detection system based on whale optimization algorithm characterized in that, The method comprises the following steps: a data acquisition module is configured to acquire human behavior data, wherein the behavior data is composed of three-axis acceleration, angular velocity and angle value; a feature selection module is configured to extract time domain features of the human behavior data, convert the extracted time domain features into frequency domain features, divide the frequency domain features into high frequency signals and low frequency signals, and use three-layer wavelet packet transform to mine the hidden features in the high frequency signals and the low frequency signals, and extract the third layer node energy spectrum of the wavelet packet as time-frequency domain features; an improved whale optimization algorithm is used to remove redundant features in the initial time domain, frequency domain and time-frequency domain features, and perform feature selection. (1) Set the initial parameters of the whale population, iterate, calculate the fitness value of each individual, and update the global optimal individual; (2) According to the adaptive multi-population merging strategy, the population is divided into multiple sub-populations, the number of sub-populations is determined; The composition of each sub-population is determined, and the optimal individual of the sub-population is updated, considering the distance and fitness value between individuals, a new sub-population individual selection mechanism is formulated, and according to the normalized individual fitness value and the distance between the individual and the global optimal individual, the comprehensive value of each individual is calculated; (3) A random recombination mechanism is introduced to promote information exchange between sub-populations, and individuals will be randomly assigned to different sub-populations; (4) Each sub-population updates the individual position in turn; According to the iteration completion condition, the update is completed, and the final obtained feature is obtained; The detection module is used to input the selected features into the SVM model, and output the recognition detection result of whether the falling behavior occurs.

8. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium is used to store computer instructions, which are executed by the processor to implement the human fall detection method based on the improved whale algorithm as claimed in any one of claims 1-6.

9. An electronic device, comprising: It comprises: A processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the human fall detection method based on the improved whale algorithm as claimed in any one of claims 1-6.

Images