Fall detection method and system based on variant particle swarm optimization algorithm and convolutional neural network

Abstract

The present invention discloses a fall detection method and system using a variant particle swarm algorithm and convolutional neural network. The method comprises a video acquisition module, a central processing module, a cloud server, and a user terminal. Furthermore, the system includes a fall detection method using a variant particle swarm algorithm and convolutional neural network. The video acquisition module and the central processing module are connected via wires, the central processing module is communicatively connected to the cloud server, and the cloud server is communicatively connected to the user terminal. The video acquisition module uses a network camera to capture video images of the monitored area and transmits the video images to the central processing module. This method improves the prediction accuracy and convergence speed of the fall model without the need for wearable devices, at low cost, and without increasing the model size. It also addresses issues such as false positives, missed positives, and long training times associated with fall detection models.

Description

Technical Field

[0001] The present invention belongs to the technical field of human fall detection and identification, and specifically relates to a fall detection method and system based on a variant particle swarm algorithm and a convolutional neural network. Background Art

[0002] Currently, falls have become the leading cause of death from injuries among people over 65 in my country. Promptly detecting a fall and sending a distress signal not only buys the injured person the best time for treatment but also prevents potential secondary injuries. Therefore, implementing effective fall detection technology is crucial.

[0003] Fall detection methods can be divided into wearable sensor-based, environment-based, and video image-based detection methods. Wearable sensor-based fall detection methods require users to wear the device for a long time, which may cause discomfort, frequent charging, forgetting to wear the device, and other problems; environment-based detection methods have high equipment costs and are easily interfered with by sound and light signals in the environment, resulting in a high misdetection rate; video image-based detection methods do not require wearable devices and are cheap, but their model structures are complex and require long computational times. Methods for fall detection based on video images, such as Publication No. CN113435306A, use a hybrid cascade convolution fall detection method to fully utilize the information extracted by the spatiotemporal information integration module for detection; Publication No. CN114120370A uses convolutional neural networks and long short-term memory networks to improve the accuracy of human fall detection, but increasing the model size often leads to a significant increase in model computation time, which is not conducive to transplantation and application.

[0004] Therefore, it is of great research value and practical significance to study a human fall behavior detection method that does not require users to wear clothing, has high recognition accuracy, and is highly portable. Summary of the Invention

[0005] The purpose of the present invention is to provide a fall detection method and system based on a variant particle swarm algorithm and a convolutional neural network, which improves the prediction accuracy and convergence speed of the fall model without the need for wearable equipment, low cost, and without increasing the model scale, and solves problems such as false positives, missed positives, and long training time of the fall detection model, so as to solve the problems in the prior art raised in the above background technology.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The fall detection method based on the variant particle swarm algorithm and convolutional neural network includes the following steps:

[0008] S1. Establishing a fall detection model based on a convolutional neural network. Step S1 includes:

[0009] Step S11: Acquire and annotate a human fall behavior dataset;

[0010] Step S12: Select and fine-tune the convolutional neural network model;

[0011] S2. Optimize the parameters of the fall detection model based on convolutional neural network using the mutation particle swarm algorithm;

[0012] S3, assigning the parameters corresponding to the global optimal solution to the fall detection model based on convolutional neural network;

[0013] S4, iteratively train the initialized convolutional neural network to obtain a trained convolutional neural network model;

[0014] S5. Verify the fall detection model based on the mutant particle swarm algorithm and convolutional neural network. Step S5 includes:

[0015] Step S51: Determine whether the fall detection model based on the mutant particle swarm algorithm and convolutional neural network reaches the specified accuracy. If not, jump to step S2;

[0016] Step S52: Save the model.

[0017] Preferably, the specific steps of S2 are as follows:

[0018] Step S21: determining the fitness function of the particle swarm according to the loss function of the convolutional neural network;

[0019] Step S22: encoding and sorting the weights and thresholds of the convolutional neural network according to its topological structure;

[0020] Step S23: Determine the dimension of the particles according to the encoding length, and initialize the number of particle swarms, inertia weight, learning parameters and maximum number of iterations;

[0021] Step S24: Initialize the random particle swarm using its initial position, initial velocity, individual optimal position, and global optimal position. The initialization method is as follows:

[0022] x i =(x max -x min )*rand+x min

[0023] v i =(v max -v min )*rand+v min

[0024] Where: v i and x i are the velocity and position of the i-th particle, xmax and x min Indicates the maximum and minimum values ​​of the position, v max and v min Indicates the maximum and minimum speed values.

[0025] Step S25: Determine whether the mutation condition is met; if not, jump to step S2.7;

[0026] Step S26: Use random numbers to mutate particles whose cosine similarity is greater than the average cosine similarity. The schematic diagram of the cosine similarity mutation strategy is as follows: Figure 1 As shown; the calculation formulas for cosine similarity and average cosine similarity are as follows:

[0027]

[0028]

[0029] Where: gbest is the global optimal solution of the current iteration, x i is the position of the i-th particle.

[0030] Step S27: Update the speed and position of the particle of the current iteration according to the speed and position of the particle of the previous generation; the speed and position iterative update formula can be described as:

[0031]

[0032]

[0033] Where N is the initial number of particles, and The velocity and position of the i-th particle at the k+1th iteration are measured respectively. ω represents the inertia weight, which reflects the influence of the previous velocity on the current velocity. c1 and c2 represent the acceleration factors, which are usually represented by two real numbers. r1 and r2 are two random numbers in the interval (0,1).

[0034] Step S28: obtaining the fitness value of the current iteration according to the particle position of the current iteration;

[0035] Step S29: Determine the individual optimal position of the current iteration and the global optimal position;

[0036] Step S210: Determine whether the current iteration has reached the maximum number of iterations or the corresponding accuracy requirement; if not, jump to step S25; if so, output the global optimal position.

[0037] Preferably, in S4, the iterative formula is as follows:

[0038]

[0039]

[0040] Where α is the learning rate, which is usually a constant between 0.01 and 0.001.

[0041] The fall detection system of the variant particle swarm algorithm and convolutional neural network includes a video acquisition module, a central processing module, a cloud server, and a user end, and also includes a fall detection method of the variant particle swarm algorithm and convolutional neural network. The video acquisition module is connected to the central processing module through a wire, the central processing module is communicatively connected to the cloud server, the cloud server is communicatively connected to the user end, and the video acquisition module uses a network camera to capture video images of the monitored area and transmits the video images to the central processing module.

[0042] Preferably, the central processing module includes an information storage unit, an information processing unit, and an alarm unit. The information storage unit stores relevant programs, algorithm models, and warning information; the information processing unit processes the image data of the monitoring area collected by the video acquisition module and calls the algorithm model of the information storage unit to detect fall information; the alarm unit includes an information pop-up box alarm on the user-side APP and an external sound and light alarm.

[0043] Preferably, the central processing module deploys a convolutional neural network algorithm based on variant particle swarm optimization. After receiving the video image data, the central processing module will process the fall information. If a fall occurs, a distress signal will be sent locally and the detected fall image, fall time and other data will be saved. At the same time, the relevant data will be uploaded to the cloud server, and the cloud server will send the data to the user end. The user can set the distress signal sensitivity, receive and view the fall information through the mobile phone APP, and take relevant measures to rush to the scene for treatment and assistance.

[0044] Preferably, the operation steps of the system include data initialization, data collection, data processing, data uploading, and data sending, wherein the data initialization includes the method for setting the sensitivity of the distress signal proposed by the present invention;

[0045] Data collection includes multiple network cameras installed by the user to collect image data of the monitored area in real time;

[0046] Data processing includes a convolutional neural network algorithm model based on variant particle swarm optimization deployed in the central processing module. The algorithm model determines whether a fall has occurred in the received image information. If not, the algorithm returns to the data collection step. If so, the algorithm performs a sensitivity threshold judgment. If the fall time does not exceed the preset sensitivity threshold, the algorithm re-judgments whether a fall has occurred. If the fall time exceeds the preset sensitivity threshold, the algorithm enters the data upload step.

[0047] Data uploading includes uploading relevant information such as fall images and fall time to the cloud server. The fall image information stored locally is used for iterative upgrades of the algorithm model;

[0048] Data distribution includes pushing the fall image information stored in the cloud server to the user end. Users can check the fall information through the mobile phone APP and conduct secondary confirmation of the fall information to eliminate low-probability events such as accidental triggering.

[0049] Technical effects and advantages of the present invention: Compared with the existing technology, the fall detection method and system based on the variant particle swarm algorithm and convolutional neural network proposed in the present invention have the following advantages:

[0050] 1. First, the present invention uses a particle swarm optimization algorithm to optimize the parameters of the convolutional neural network. Compared with other hybrid models, this model does not require additional model size, improving the accuracy of fall detection while ensuring the model's computational speed. Compared with traditional convolutional neural network models, this model has better optimization capabilities, improving the model's convergence speed and the accuracy of fall detection. Secondly, the present invention adopts a particle swarm mutation mechanism, using cosine similarity to evaluate the degree of particle aggregation, causing particles with large aggregation to mutate, solving the problem of premature convergence in the standard particle swarm algorithm and improving the particle swarm's population diversity and global search capability. Finally, this paper proposes a method for implementing a real-time emergency call system for fall detection based on video images. The video stream collected by the video acquisition device is transmitted to the control center, where a convolutional neural network based on mutation particle swarm optimization is used for fall detection. If a fall is detected, a distress signal is sent locally and relevant data such as the detected fall image and fall time are sent to a cloud server. The cloud server transmits this data to a mobile phone app. Security personnel can view the fall situation through the mobile app and take appropriate measures to rush to the scene for treatment and assistance.

[0051] 2. This method improves the accuracy of fall behavior recognition and the convergence speed of the model without increasing the scale of the convolutional neural network model, and proposes an implementation method of a real-time help-seeking system for fall detection based on video images, which can effectively help the injured get timely assistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Schematic diagram of the cosine similarity-based mutation strategy of the present invention;

[0053] Figure 2 This is a framework diagram of the convolutional neural network recognition module of the present invention;

[0054] Figure 3 This is an algorithm flow chart of the convolutional neural network optimized based on the variant particle swarm algorithm of the present invention;

[0055] Figure 4 This is a structural diagram of the fall assistance system based on the convolutional neural network of the variant particle swarm of the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0057] The present invention provides Figure 1-4 The fall detection method using the variant particle swarm algorithm and convolutional neural network shown in the figure includes the following steps:

[0058] S1. Establish a fall detection model based on a convolutional neural network. Step S1 includes:

[0059] Step S11: Acquire and annotate a human fall behavior dataset;

[0060] Step S12: Select and fine-tune the convolutional neural network model;

[0061] S2. Optimize the parameters of the fall detection model based on convolutional neural network using the mutation particle swarm algorithm;

[0062] S3, assigning the parameters corresponding to the global optimal solution to the fall detection model based on convolutional neural network;

[0063] S4, iteratively train the initialized convolutional neural network to obtain a trained convolutional neural network model;

[0064] S5. Verify the fall detection model based on the mutant particle swarm algorithm and convolutional neural network. Step S5 includes:

[0065] Step S51: Determine whether the fall detection model based on the mutant particle swarm algorithm and convolutional neural network reaches the specified accuracy. If not, jump to step S2;

[0066] Step S52: Save the model.

[0067] The specific steps of S2 are as follows:

[0068] Step S21: determining the fitness function of the particle swarm according to the loss function of the convolutional neural network;

[0069] Step S22: encoding and sorting the weights and thresholds of the convolutional neural network according to its topological structure;

[0070] Step S23: Determine the dimension of the particles according to the encoding length, and initialize the number of particle swarms, inertia weight, learning parameters and maximum number of iterations;

[0071] Step S24: Initialize the random particle swarm using its initial position, initial velocity, individual optimal position, and global optimal position. The initialization method is as follows:

[0072] x i =(x max -x min )*rand+x min

[0073] v i =(v max -v min )*rand+v min

[0074] Where: v i and x i are the velocity and position of the i-th particle, x max and x min Indicates the maximum and minimum values ​​of the position, v max and v min Indicates the maximum and minimum speed values.

[0075] Step S25: Determine whether the mutation condition is met; if not, jump to step S2.7;

[0076] Step S26: Use random numbers to mutate particles whose cosine similarity is greater than the average cosine similarity. The schematic diagram of the cosine similarity mutation strategy is as follows: Figure 1 As shown; the calculation formulas for cosine similarity and average cosine similarity are as follows:

[0077]

[0078]

[0079] Where: gbest is the global optimal solution of the current iteration, x i is the position of the i-th particle.

[0080] Step S27: Update the speed and position of the particle of the current iteration according to the speed and position of the particle of the previous generation; the speed and position iterative update formula can be described as:

[0081]

[0082]

[0083] Where N is the initial number of particles, and The velocity and position of the i-th particle at the k+1th iteration are measured respectively. ω represents the inertia weight, which reflects the influence of the previous velocity on the current velocity. c1 and c2 represent the acceleration factors, which are usually represented by two real numbers. r1 and r2 are two random numbers in the interval (0,1).

[0084] Step S28: Obtain the fitness value of the current iteration according to the particle position of the current iteration;

[0085] Step S29: Determine the individual optimal position of the current iteration and the global optimal position;

[0086] Step S210: Determine whether the current iteration has reached the maximum number of iterations or the corresponding accuracy requirement; if not, jump to step S25; if so, output the global optimal position.

[0087] In S4, the iterative formula is as follows:

[0088]

[0089]

[0090] Where α is the learning rate, which is usually a constant between 0.01 and 0.001.

[0091] The fall detection system of the mutant particle swarm algorithm and convolutional neural network includes a video acquisition module, a central processing module, a cloud server, and a user end, and also includes a fall detection method of the mutant particle swarm algorithm and convolutional neural network. The video acquisition module and the central processing module are connected by a wire, the central processing module is communicated with the cloud server, the cloud server and the user end are communicated, and the video acquisition module uses a network camera to capture video images of the monitored area and transmits the video images to the central processing module.

[0092] The central processing module includes an information storage unit, an information processing unit, and an alarm unit. The information storage unit stores relevant programs, algorithm models, and warning information. The information processing unit processes the image data of the monitoring area collected by the video acquisition module and uses the algorithm model of the information storage unit to detect falls. The alarm unit includes a pop-up alarm in the user-side app and an external sound and light alarm.

[0093] The central processing module deploys a convolutional neural network algorithm based on variant particle swarm optimization. After receiving the video image data, it will process the fall information in the central processing module. If a fall occurs, a distress signal will be sent locally and the detected fall image, fall time and other data will be saved. At the same time, the relevant data will be uploaded to the cloud server, and the cloud server will send the data to the user end. The user can set the distress signal sensitivity, receive and view fall information, and take relevant measures to rush to the scene for treatment and assistance through the mobile phone APP.

[0094] The operation steps of this system include data initialization, data acquisition, data processing, data upload, and data distribution, among which data initialization includes the method for setting the sensitivity of the distress signal proposed in the present invention; data acquisition includes real-time image data of the monitored area collected by multiple network cameras installed by the user; data processing includes a convolutional neural network algorithm model based on variant particle swarm optimization deployed in the central processing module, which determines whether a fall has occurred in the received image information. If not, the system returns to the data acquisition step; if so, a sensitivity threshold judgment is performed; if the fall time does not exceed the preset sensitivity threshold, a re-judgment is made as to whether a fall has occurred; if the fall time exceeds the preset sensitivity threshold, the system enters the data upload step; data uploading includes uploading relevant information such as the fall image and the fall time to the cloud server, wherein the fall image information stored locally is used for iterative upgrades of the algorithm model; data distribution includes pushing the fall image information stored on the cloud server to the user end in the background, and the user can check the fall information through the mobile phone APP and conduct a secondary confirmation of the fall information to eliminate low-probability events such as accidental triggering.

[0095] Example:

[0096] The fall detection method based on the variant particle swarm algorithm and convolutional neural network includes the following steps:

[0097] Step S1: Acquire and annotate the human fall behavior dataset

[0098] A total of 7782 fall image data were collected using web crawlers and fall experiments. After excluding some unusable data, the number of valid data was 7713. The fall dataset was then divided into a training set and a test set in a ratio of 9:1.

[0099] Step S2: Figure 2 As shown, a fall detection model based on convolutional neural network is established

[0100] The fall detection model based on convolutional neural networks consists of two parts: an extraction module and a recognition module. The extraction module uses VGG19 to normalize the human fall detection behavior dataset to obtain a feature map, and then uses the OpenPose model to extract human joint information from the feature map. The recognition module uses a convolutional neural network. The convolutional neural network structure set by the present invention includes three convolutional layers with convolution kernel sizes of 5×5, 3×3, and 3×3, respectively, and the number of convolution filters is 32, 64, and 64, respectively. The step size is 1, the image padding uses the same, and the activation function uses Relu. Three pooling layers all use maximum pooling layers, the kernel size of the pooling layer is 2×2, and the number of filters is 32, 64, and 64, respectively. The exit layer probability parameter is set to 0.5.

[0101] Step S3: Determine and solve the fitness function of the particle swarm algorithm, use the particle swarm algorithm to optimize the weights and thresholds of the neural network, and use the mutation particle swarm algorithm to optimize the parameters of the fall detection model.

[0102] The loss function of the convolutional neural network model is used as the fitness evaluation function of the mutation particle swarm algorithm. The number of parameters of the convolutional neural network of the recognition module in step S2 is used as the dimension of the particle. The number of particles is set to 100, the particle position range is [-0.02, 0.02], and the particle speed range is [-0.005, 0.005]. The initial position and speed are randomly generated within the position and speed range. The probability of particle swarm mutation decreases linearly from 50% to 10% with the number of iterations. The iterative termination condition is that the accuracy meets the requirements or the iteration reaches the maximum number of iterations. The optimization algorithm flow chart is as follows: Figure 3 shown.

[0103] In step S4, the parameters corresponding to the global optimal solution are assigned to the fall detection model based on the convolutional neural network and then iterative training is performed to obtain a trained convolutional neural network model.

[0104] The global optimal solution output by the mutated particle swarm algorithm was used as the initial parameters for the convolutional neural network model in the recognition module. The stochastic gradient descent algorithm was then used for further iterative training to obtain the fall detection model. During the training phase, the number of training iterations was set to 100, and the learning rate was set to 0.001. Based on the test set's detection results, the model with the highest accuracy was selected as the final output model.

[0105] Step S5: Transplant the fall detection model based on the variant particle swarm algorithm and convolutional neural network into the fall assistance system. The system structure diagram is as follows: Figure 4 shown.

[0106] Data collection: The network camera of the video surveillance system collects the video stream.

[0107] Data processing: The collected video stream is analyzed and detected on the embedded device. If a fall is detected, the sensitivity threshold is judged. If the fall time does not exceed the preset sensitivity threshold, it is re-judged whether a fall has occurred. If the fall time exceeds the preset sensitivity threshold, a distress signal is immediately sent out on the sound and light alarm and a message pop-up box is sent out on the user-side APP.

[0108] Data upload and download: The embedded device transmits the fall image, fall time and other information to the cloud server for storage, and the cloud server then pushes the data to the user terminal APP.

[0109] User terminal: Security personnel receive the specific time and situation of the fall on their mobile phones, and take appropriate measures to seek help in a timely manner.

[0110] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A fall detection method using a variant particle swarm algorithm and a convolutional neural network comprises the following steps, characterized in that: S1. Establishing a fall detection model based on a convolutional neural network. Step S1 includes: Step S11: Acquire and annotate a human fall behavior dataset; Step S12: Select and fine-tune the convolutional neural network model; S2. Optimize the parameters of the fall detection model based on convolutional neural network using the mutation particle swarm algorithm; S3, assigning the parameters corresponding to the global optimal solution to the fall detection model based on convolutional neural network; S4, iteratively train the initialized convolutional neural network to obtain a trained convolutional neural network model; S5. Verify the fall detection model based on the mutant particle swarm algorithm and convolutional neural network. Step S5 includes: Step S51: Determine whether the fall detection model based on the variant particle swarm algorithm and convolutional neural network reaches the specified accuracy. If not, jump to step S2; Step S52: save the model; The specific steps of S2 are as follows: Step S21: determining the fitness function of the particle swarm according to the loss function of the convolutional neural network; Step S22: encoding and sorting the weights and thresholds of the convolutional neural network according to its topological structure; Step S23: Determine the dimension of the particles according to the encoding length, and initialize the number of particle swarms, inertia weight, learning parameters and maximum number of iterations; Step S24: Initialize the random particle swarm using its initial position, initial velocity, individual optimal position, and global optimal position. The initialization method is as follows: Where: and are the velocity and position of the i-th particle, and Indicates the maximum and minimum values ​​of the position, and Indicates the maximum and minimum values ​​of speed; Step S25: Determine whether the mutation condition is met; if not, jump to step S2.7; Step S26: Use random numbers to mutate particles whose cosine similarity is greater than the average cosine similarity; the calculation formulas for cosine similarity and average cosine similarity are as follows: Where: is the global optimal solution of the current iteration, is the position of the i-th particle; Step S27: Update the speed and position of the particle of the current iteration according to the speed and position of the particle of the previous generation; the speed and position iteration update formula is described as: In the formula is the initial number of particles, and Measure the velocity and position of the i-th particle at the k+1-th iteration respectively; Indicates the inertia weight, which reflects the impact of the previous speed on the current speed. and The acceleration factor is usually expressed as two real numbers, and are two random numbers in the interval (0,1); Step S28: obtaining the fitness value of the current iteration according to the particle position of the current iteration; Step S29: Determine the individual optimal position of the current iteration and the global optimal position; Step S210: Determine whether the current iteration has reached the maximum number of iterations or the corresponding accuracy requirement; if not, jump to step S25; if so, output the global optimal position.

2. The fall detection method using a variant particle swarm algorithm and a convolutional neural network according to claim 1 is characterized by: In the S4, the iterative formula is as follows: In the formula is the learning rate, usually a constant between 0.01 and 0.

001.

3. A fall detection system using a variant particle swarm algorithm and a convolutional neural network, configured to implement the fall detection method using a variant particle swarm algorithm and a convolutional neural network according to claim 1 or 2, comprising a video acquisition module, a central processing module, a cloud server, and a user terminal, and also comprising the fall detection method using a variant particle swarm algorithm and a convolutional neural network, characterized in that: The video acquisition module is connected to the central processing module through a wire, the central processing module is connected to the cloud server, the cloud server is connected to the user end, and the video acquisition module uses a network camera to capture video images of the monitored area and transmit the video images to the central processing module.

4. The fall detection system using a variant particle swarm algorithm and a convolutional neural network according to claim 3 is characterized by: The central processing module includes an information storage unit, an information processing unit, and an alarm unit. The information storage unit stores relevant programs, algorithm models, and early warning information; the information processing unit processes the image data of the monitoring area collected by the video acquisition module, and calls the algorithm model of the information storage unit to detect fall information; the alarm unit includes an information pop-up box alarm of the user-side APP and an external sound and light alarm.

5. The fall detection system using a variant particle swarm algorithm and a convolutional neural network according to claim 3 is characterized by: The central processing module deploys a convolutional neural network algorithm based on variant particle swarm optimization. After receiving the video image data, it will process the fall information in the central processing module. If a fall occurs, a distress signal will be sent locally and the detected fall image, fall time and other data will be saved. At the same time, the relevant data will be uploaded to the cloud server. The cloud server will send the data to the user end. The user can set the distress signal sensitivity, receive and view the fall information through the mobile phone APP, and take relevant measures to rush to the scene for treatment and assistance.

6. The fall detection system using a variant particle swarm algorithm and a convolutional neural network according to claim 4, characterized in that: The operation steps of this system include data initialization, data collection, data processing, data uploading, and data sending. Among them, data initialization includes the method of setting the sensitivity of the distress signal proposed by the present invention; Data collection includes multiple network cameras installed by the user to collect image data of the monitored area in real time; Data processing includes a convolutional neural network algorithm model based on variant particle swarm optimization deployed in the central processing module. The algorithm model determines whether a fall has occurred in the received image information. If not, the algorithm returns to the data collection step. If so, the algorithm performs a sensitivity threshold judgment. If the fall time does not exceed the preset sensitivity threshold, the algorithm re-judgments whether a fall has occurred. If the fall time exceeds the preset sensitivity threshold, the algorithm enters the data upload step. Data uploading includes uploading relevant information such as fall images and fall time to the cloud server. The fall image information stored locally is used for iterative upgrades of the algorithm model; Data distribution includes pushing the fall image information stored in the cloud server to the user end. The user checks the fall information through the mobile phone APP and reconfirms the fall information to eliminate low-probability events such as accidental triggering.

Images