Fall detection method and system, electronic device and medium

Abstract

The present invention relates to the technical field of human health monitoring. Disclosed are a fall detection method and system, an electronic device and a medium. The method comprises: acquiring an image for detection at a current moment; using a lightweight YOLOv5s network to process the image for detection at the current moment, so as to determine whether a target in the image for detection at the current moment has fallen, thereby obtaining a first determination result; if the first determination result is yes, using OpenPose to process the image for detection at the current moment, so as to determine whether the target in the image for detection at the current moment has fallen, thereby obtaining a second determination result; if the second determination result is yes, determining that the target in the image for detection at the current moment has fallen; and, if the second determination result or the first determination result is no, updating the current moment, and returning to the step of acquiring an image for detection at a current moment. The present invention can reduce occupation of computational resources of edge devices such as Raspberry Pi, thus greatly improving the device reliability.

Description

Fall detection method, system, electronic device and medium

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on February 23, 2024, with application number 202410202746.4 and invention name “A fall detection method, system, electronic device and medium”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present invention relates to the technical field of human health monitoring, and in particular to a fall detection method, system, electronic equipment and medium. Background Art

[0003] Traditional fall detection deployment methods simply streamline the model, but these safety detection devices require constant operation. Due to the limited computing power of edge devices like the Raspberry Pi, they sometimes can't run programs uninterrupted for extended periods, potentially leading to system freezes and lags. Detecting falls among the elderly, children, and people with disabilities is highly time-sensitive. If an incident occurs, the detection system may be unable to signal the caregiver due to freezes or other issues, potentially missing the optimal opportunity for rescue.

[0004] Summary of the Invention

[0005] The purpose of the present invention is to provide a fall detection method, system, electronic device and medium, which can reduce the computing resources occupied by edge devices such as Raspberry Pi, greatly enhance the reliability of the device, and thus improve the timeliness of fall detection.

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

[0007] A fall detection method, comprising:

[0008] Get the image to be detected at the current moment;

[0009] A lightweight YOLOv5s network is used to process the image to be detected at the current moment to determine whether the target in the image to be detected at the current moment has fallen, and a first judgment result is obtained;

[0010] If the first judgment result is yes, use OpenPose to process the image to be detected at the current moment to determine whether the target in the image to be detected at the current moment has fallen, and obtain a second judgment result;

[0011] If the second judgment result is yes, confirming that the target in the image to be detected at the current moment has fallen;

[0012] If the second judgment result is no, then update the current moment and return to the step of obtaining the image to be detected at the current moment;

[0013] If the first judgment result is no, the current moment is updated and the process returns to the step of obtaining the image to be detected at the current moment.

[0014] Optionally, the lightweight YOLOv5s network includes: SPP, a first Ghost Bottleneck, a second Ghost Bottleneck, a third Ghost Bottleneck, a fourth Ghost Bottleneck, a fifth Ghost Bottleneck, a sixth Ghost Bottleneck, a first upsample, a second upsample, a first concat, a second concat, a third concat, a fourth concat, a first conv, a second conv, a third conv, a fourth conv, and a fifth conv;

[0015] The output end of the first Ghost Bottleneck is connected to the input end of the second Ghost Bottleneck and the input end of the first concat respectively; the output end of the second Ghost Bottleneck is connected to the input end of the SPP and the input end of the second concat respectively; the output end of the SPP is connected to the input end of the fourth concat and the input end of the second upsample respectively;

[0016] The output end of the second upsample is connected to the input end of the second concat, the output end of the second concat is connected to the input end of the fourth Ghost Bottleneck, the output end of the fourth Ghost Bottleneck is connected to the input end of the first upsample and the input end of the third concat respectively, the output end of the first upsample is connected to the input end of the first concat, the output end of the first concat is connected to the input end of the third Ghost Bottleneck, and the output end of the third Ghost Bottleneck is connected to the input end of the first conv and the input end of the third conv respectively;

[0017] The output end of the first conv is connected to the input end of the third concat, the output end of the third concat is connected to the input end of the fifth Ghost Bottleneck, the output end of the fifth Ghost Bottleneck is connected to the input end of the second conv and the input end of the fourth conv respectively, the output end of the second conv is connected to the input end of the fourth concat, and the output end of the fourth concat is connected to the input end of the sixth Ghost Bottleneck;

[0018] The output end of the sixth Ghost Bottleneck is connected to the input end of the fifth conv.

[0019] Optionally, if the second judgment result is yes, after the step of confirming that the target in the image to be detected at the current moment has fallen, the method further includes: issuing an alarm.

[0020] A fall detection system comprising:

[0021] An acquisition module is used to obtain the image to be detected at the current moment;

[0022] A first judgment module is used to process the image to be detected at the current moment using a lightweight YOLOv5s network to determine whether the target in the image to be detected at the current moment has fallen, and obtain a first judgment result;

[0023] A second judgment module is configured to, if the first judgment result is yes, use OpenPose to process the image to be detected at the current moment, determine whether the target in the image to be detected at the current moment has fallen, and obtain a second judgment result;

[0024] an alarm module, configured to confirm that the target in the image to be detected at the current moment has fallen down if the second judgment result is yes;

[0025] A first returning module, configured to update the current moment and return to the step of obtaining the image to be detected at the current moment if the second judgment result is no;

[0026] The second returning module is used to update the current moment and return to the step of obtaining the image to be detected at the current moment if the first judgment result is no.

[0027] Optionally, the lightweight YOLOv5s network includes: SPP, a first Ghost Bottleneck, a second Ghost Bottleneck, a third Ghost Bottleneck, a fourth Ghost Bottleneck, a fifth Ghost Bottleneck, a sixth Ghost Bottleneck, a first upsample, a second upsample, a first concat, a second concat, a third concat, a fourth concat, a first conv, a second conv, a third conv, a fourth conv, and a fifth conv;

[0028] The output end of the first Ghost Bottleneck is connected to the input end of the second Ghost Bottleneck and the input end of the first concat respectively; the output end of the second Ghost Bottleneck is connected to the input end of the SPP and the input end of the second concat respectively; the output end of the SPP is connected to the input end of the fourth concat and the input end of the second upsample respectively;

[0029] The output end of the second upsample is connected to the input end of the second concat, the output end of the second concat is connected to the input end of the fourth Ghost Bottleneck, the output end of the fourth Ghost Bottleneck is connected to the input end of the first upsample and the input end of the third concat respectively, the output end of the first upsample is connected to the input end of the first concat, the output end of the first concat is connected to the input end of the third Ghost Bottleneck, and the output end of the third Ghost Bottleneck is connected to the input end of the first conv and the input end of the third conv respectively;

[0030] The output end of the first conv is connected to the input end of the third concat, the output end of the third concat is connected to the input end of the fifth Ghost Bottleneck, the output end of the fifth Ghost Bottleneck is connected to the input end of the second conv and the input end of the fourth conv respectively, the output end of the second conv is connected to the input end of the fourth concat, and the output end of the fourth concat is connected to the input end of the sixth Ghost Bottleneck;

[0031] The output end of the sixth Ghost Bottleneck is connected to the input end of the fifth conv.

[0032] Optionally, the fall detection system further includes: an alarm module, configured to issue an alarm.

[0033] An electronic device, comprising:

[0034] A memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the above-mentioned fall detection method.

[0035] A computer-readable storage medium stores a computer program, which implements the above-mentioned fall detection method when executed by a processor.

[0036] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0037] The present invention uses a lightweight YOLOv5s network to process the image to be detected at the current moment to determine whether the target in the image to be detected at the current moment has fallen, and obtains a first judgment result; if the first judgment result is yes, OpenPose is used to process the image to be detected at the current moment to determine whether the target in the image to be detected at the current moment has fallen, and obtains a second judgment result. The lightweight YOLOv5s network is first used to detect the image, and then OpenPose is used to perform a second detection. A nested structure is used instead of concurrent operation, which can reduce the computing resources occupied by edge devices such as Raspberry Pi, greatly enhance the reliability of the device, and thereby improve the timeliness of detecting the fall. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 is a schematic diagram of the structure of the YOLOv5s network;

[0040] FIG2 is a schematic diagram of the structure of Ghost Bottleneck; FIG2 (a) is the first Ghost module in Ghost Bottleneck, and FIG2 (b) is the second Ghost module in Ghost Bottleneck;

[0041] FIG3 is a schematic diagram of the structure of the lightweight YOLOv5s network provided by the present invention;

[0042] FIG4 is a flow chart of the fall detection method provided by the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] In order to make the objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Existing fall detection algorithms ignore the results of long-term testing on the device. The basic idea of ​​this invention is to first lightweight the YOLOv5s model and then, after verifying the lightweight results, embed them into an even lighter network model (OpenPose). The YOLOv5s model is first used to determine whether a person has fallen. If the result is yes, OpenPose is further used to determine whether the person has fallen.

[0046] As shown in FIG4 , the embodiment of the present invention provides a fall detection method for the elderly, children, and people with disabilities. This method can be deployed on a Raspberry Pi and includes:

[0047] Get the image to be detected at the current moment (camera acquisition).

[0048] The lightweight YOLOv5s network is used to process the image to be detected at the current moment to determine whether the target (person) in the image to be detected at the current moment has fallen, and a first judgment result is obtained.

[0049] If the first judgment result is yes, OpenPose is used to process the image to be detected at the current moment to determine whether the target in the image to be detected at the current moment has fallen, and obtain a second judgment result.

[0050] If the second judgment result is yes, it is confirmed that the target in the image to be detected at the current moment has fallen.

[0051] If the second judgment result is no, the current moment is updated, and the process returns to the step of obtaining the image to be detected at the current moment.

[0052] If the first judgment result is no, the current moment is updated and the process returns to the step of obtaining the image to be detected at the current moment.

[0053] In practical applications, if the second judgment result is yes, after the step of confirming that the target in the image to be detected at the current moment has fallen, the method further includes: issuing an alarm.

[0054] As shown in Figure 1, the convolutional layers of the Conv and C3 modules of YOLOv5s exhibit feature map redundancy. The Conv modules in the YOLOv5s network introduce residual connections between repeated convolutional layers. Because shallow feature maps are repeatedly calculated and fused in subsequent convolutional layers using residual connections, this can lead to redundant information between shallow and deep feature maps. The settings of parameters such as the stride and kernel size of the convolutional layers can cause the receptive fields of different convolutional layers to overlap, resulting in feature map redundancy. For example, convolution operations with large strides can lead to redundancy between adjacent pixels in feature maps. In the Conv and C3 modules, the use of smaller kernels to reduce the number of channels during dimensionality reduction and the subsequent expansion through standard convolution operations can lead to a certain degree of redundancy between channels in feature maps. The structure of the Ghost bottleneck is similar to the residual component in ResNet. The Ghost Bottleneck architecture consists of two Ghost modules. Wherein, the first Ghost module is shown in part (a) of Figure 2, and its role is to increase the number of channels of the input feature map and to perform expansion processing for subsequent operations, while the second Ghost module is shown in part (b) of Figure 2, and its role is to reduce the number of channels of the output feature map so that it matches the diameter structure in the network, and to connect the information transmission of the two Ghost modules through the diameter structure. As can be seen from Figure 2, the difference between the two Ghost modules is that the Relu activation function is used after the first Ghost, while batch normalization is used in each layer of the latter. By such a structural mode, the model can effectively reduce model parameters and computational complexity, and can also optimize the feature map through the Ghost module to improve the detection efficiency of the model. Therefore, in the present invention, in order to optimize the computational loss of the YOLOv5s network, this Ghost bottleneck will be used to optimize the YOLOv5s network to improve the robustness of the network in actual detection. Through this process, a more compact and useful feature representation can be constructed to improve the performance of tasks such as detection or classification.

[0055] Ghost Bottleneck increases the width of the network by dividing the input feature maps into two groups. Specifically, Ghost Bottleneck divides the input feature maps into a main branch and a secondary branch. The main branch is responsible for extracting primary features, while the secondary branch is responsible for extracting auxiliary features. This grouping operation effectively increases the width of the network and improves the expressive power of the model. Ghost Bottleneck groups the input feature maps. To reduce computational complexity, Ghost Bottleneck introduces an additional linear projection layer, which is used to project the feature maps of the secondary branches to the same dimension as the main branch. This reduces the computational workload of the secondary branches and keeps the feature dimensions of the main and secondary branches consistent, facilitating subsequent feature fusion. This grouping operation effectively increases the width of the network, thereby improving the expressive power of the model. Furthermore, Ghost Bottleneck employs a residual connection method, which adds the feature maps of the main branch to the feature maps of the secondary branches. This residual connection method helps the model better learn feature representations and accelerates network convergence. Through the residual connection, the model can better transfer the gradient, thereby improving the training effect of the model, helping the model to better learn feature representation, and speeding up the convergence speed of the network. Therefore, the present invention adopts Ghost Bottleneck to replace the Bottleneck CSP method in the original YOLO network backbone to optimize the network parameters and calculation scale. At the same time, the standard convolution of part of the original YOLOv5s network is replaced with a more lightweight decomposition convolution structure after pre-training, which greatly optimizes the network parameter scale and computing resource consumption without affecting the network pre-training detection accuracy. However, it is worth noting that since the convolution structures at the beginning and end of the network are used to sample the input samples and output high-dimensional classification features, the convolution structures of these two parts are not considered in the replacement process of the convolution structure, and only the middle part is processed.

[0056] Specifically, the lightweight YOLOv5s network is shown in Figure 3, including: SPP, a first Ghost Bottleneck, a second Ghost Bottleneck, a third Ghost Bottleneck, a fourth Ghost Bottleneck, a fifth Ghost Bottleneck, a sixth Ghost Bottleneck, a first upsample, a second upsample, a first concat, a second concat, a third concat, a fourth concat, a first conv, a second conv, a third conv, a fourth conv and a fifth conv; the output end of the first Ghost Bottleneck is respectively connected to the input end of the second Ghost Bottleneck and the input end of the first concat; the output end of the second Ghost Bottleneck is respectively connected to the input end of the SPP and the input end of the second concat; the output end of the SPP is respectively connected to the input end of the fourth concat and the input end of the second upsample; the output end of the second upsample is connected to the input end of the second concat, the output end of the second concat is connected to the input end of the fourth Ghost Bottleneck, and the fourth Ghost The output end of Bottleneck is respectively connected to the input end of the first upsample and the input end of the third concat, the output end of the first upsample is connected to the input end of the first concat, the output end of the first concat is connected to the input end of the third Ghost Bottleneck, and the output end of the third Ghost Bottleneck is respectively connected to the input end of the first conv and the input end of the third conv; the output end of the first conv is connected to the input end of the third concat, the output end of the third concat is connected to the input end of the fifth Ghost Bottleneck, the output end of the fifth Ghost Bottleneck is respectively connected to the input end of the second conv and the input end of the fourth conv, the output end of the second conv is connected to the input end of the fourth concat, the output end of the fourth concat is connected to the input end of the sixth Ghost Bottleneck; the output end of the sixth Ghost Bottleneck is connected to the input end of the fifth conv.

[0057] In practical applications, to address the current issues of insufficient accuracy and low efficiency in behavior monitoring, we combined OpenPose's posture recognition technology to build a human posture prediction and classification algorithm. This algorithm predicts human posture by analyzing the coordination between different limbs. When building a human posture prediction and classification algorithm, you can choose to compare different classification algorithms, such as YOLOv5n, YOLOv5x, and YOLOv5l. These algorithms are all improved versions of the YOLO (You Only Look Once) object detection algorithm. They differ in detection speed and accuracy, and the model size is gradually increasing.

[0058] Because the algorithm is optimized for edge devices, YOLOv5s was selected as the optimal model. First, the selected YOLO model performs a preliminary fall detection, displaying multiple objects in the image in real time. If a fall is initially detected, OpenPose's pose recognition technology is used to apply a convolutional neural network to each detected person to identify the locations of key points. By analyzing the relationships between key points, the person's posture and motion can be determined. These key points include the head, neck, shoulders, elbows, wrists, hips, knees, and ankles. By analyzing the relationships between key points, OpenPose can determine the person's posture and motion. In this model, the OpenPose model can be used for human pose recognition. OpenPose obtains the locations of each skeletal key point of the human body, and then uses Euclidean distance to match two skeletons to detect each individual. Missing skeletal key points in common detection can be filled in using skeletal information from the previous frame. In implementation, the OpenPose API is used to obtain key point locations and perform skeletal matching and filling operations as needed.

[0059] In accordance with the above method, the present invention further provides an embodiment of a fall detection system, the fall detection system comprising:

[0060] The acquisition module is used to obtain the image to be detected at the current moment.

[0061] The first judgment module is used to use the lightweight YOLOv5s network to process the image to be detected at the current moment to determine whether the target in the image to be detected at the current moment has fallen, and obtain a first judgment result.

[0062] The second judgment module is used to use OpenPose to process the image to be detected at the current moment if the first judgment result is yes, to determine whether the target in the image to be detected at the current moment has fallen, and obtain a second judgment result.

[0063] An alarm module is used to confirm that the target in the image to be detected at the current moment has fallen if the second judgment result is yes.

[0064] The first returning module is configured to update the current moment and return to the step of obtaining the image to be detected at the current moment if the second judgment result is no.

[0065] The second returning module is used to update the current moment and return to the step of obtaining the image to be detected at the current moment if the first judgment result is no.

[0066] As an optional implementation, the lightweight YOLOv5s network includes: SPP, a first Ghost Bottleneck, a second Ghost Bottleneck, a third Ghost Bottleneck, a fourth Ghost Bottleneck, a fifth Ghost Bottleneck, a sixth Ghost Bottleneck, a first upsample, a second upsample, a first concat, a second concat, a third concat, a fourth concat, a first conv, a second conv, a third conv, a fourth conv and a fifth conv; the output end of the first Ghost Bottleneck is respectively connected to the input end of the second Ghost Bottleneck and the input end of the first concat; the output end of the second Ghost Bottleneck is respectively connected to the input end of the SPP and the input end of the second concat; the output end of the SPP is respectively connected to the input end of the fourth concat and the input end of the second upsample; the output end of the second upsample is connected to the input end of the second concat, the output end of the second concat is connected to the input end of the fourth Ghost Bottleneck, and the fourth Ghost The output end of Bottleneck is respectively connected to the input end of the first upsample and the input end of the third concat, the output end of the first upsample is connected to the input end of the first concat, the output end of the first concat is connected to the input end of the third Ghost Bottleneck, and the output end of the third Ghost Bottleneck is respectively connected to the input end of the first conv and the input end of the third conv; the output end of the first conv is connected to the input end of the third concat, the output end of the third concat is connected to the input end of the fifth Ghost Bottleneck, the output end of the fifth Ghost Bottleneck is respectively connected to the input end of the second conv and the input end of the fourth conv, the output end of the second conv is connected to the input end of the fourth concat, the output end of the fourth concat is connected to the input end of the sixth Ghost Bottleneck; the output end of the sixth Ghost Bottleneck is connected to the input end of the fifth conv.

[0067] As an optional implementation, the fall detection system further includes: an alarm module, configured to issue an alarm.

[0068] An embodiment of the present invention further provides an electronic device, including:

[0069] A memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the fall detection method described in the above embodiment.

[0070] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the fall detection method described in the above embodiment.

[0071] The present invention has the following technical effects:

[0072] 1. This paper uses model pruning, quantization, depthwise separable convolution and other technologies to reduce the number of model parameters and computational complexity. By cropping or compressing the YOLOv5s model, the YOLOv5s model is lightweighted to reduce the model size and computational complexity, making it suitable for resource-limited devices such as the Raspberry Pi. This can improve the speed and efficiency of real-time inference while maintaining relatively high detection performance.

[0073] 2. The present invention uses OpenPose to detect human postures and can identify whether a person is standing, sitting, etc. This function is very useful for some specific scenarios and needs, such as stadiums, offices, and other places where the status of people needs to be monitored.

[0074] 3. The present invention combines lightweight models, posture detection, and real-time performance, and can achieve efficient and accurate personnel posture detection on resource-limited devices such as Raspberry Pi. This has important practical significance for many application scenarios, such as security monitoring and health assessment.

[0075] 4. The present invention ultimately achieves a technology that makes daily running programs lightweight and does not require the computing resources of edge devices such as Raspberry Pi, greatly enhancing the reliability of the device and allowing excess computing power to be used for other functional calculations, reducing hardware costs.

[0076] 5. This invention combines the YOLOv5s model with OpenPose to detect falls, making the entire system more compact and efficient, reducing processing time and resource usage. Thanks to the lightweight model and pose detection program, this project achieves high real-time performance. On resource-constrained devices like the Raspberry Pi, real-time pose detection can be performed, with prompt response and processing of the results.

[0077] 6. This invention achieves high performance and real-time performance on resource-limited devices, ensuring detection accuracy while reducing computational burden. The performance of the entire system is optimized by comprehensively considering factors such as model lightweighting, startup and inference speed.

[0078] 7. Yolov5s is a classic object detection algorithm. It is an improved version of the YOLO algorithm based on deep learning. In Yolov5s, the Ghost Bottleneck structure is widely used throughout the network to extract image features. By using Ghost Bottleneck, the present invention enables Yolov5s to better capture the detailed features of objects, improve the accuracy and generalization ability of object detection, enhance the performance and accuracy of the model, and maintain a low computational cost. In general, the use of the Ghost Bottleneck structure enables Yolov5s to achieve good results in object detection tasks, improving model performance while reducing the number of parameters and computational cost.

[0079] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0080] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A fall detection method, characterized in that: include: Get the image to be detected at the current moment; A lightweight YOLOv5s network is used to process the image to be detected at the current moment to determine whether the target in the image to be detected at the current moment has fallen, and a first judgment result is obtained; If the first judgment result is yes, use OpenPose to process the image to be detected at the current moment to determine whether the target in the image to be detected at the current moment has fallen, and obtain a second judgment result; If the second judgment result is yes, confirming that the target in the image to be detected at the current moment has fallen; If the second judgment result is no, then update the current moment and return to the step of obtaining the image to be detected at the current moment; If the first judgment result is no, the current moment is updated and the process returns to the step of obtaining the image to be detected at the current moment.

2. The fall detection method according to claim 1, wherein: The lightweight YOLOv5s network includes: SPP, a first Ghost Bottleneck, a second Ghost Bottleneck, a third Ghost Bottleneck, a fourth Ghost Bottleneck, a fifth Ghost Bottleneck, a sixth Ghost Bottleneck, a first upsample, a second upsample, a first concat, a second concat, a third concat, a fourth concat, a first conv, a second conv, a third conv, a fourth conv and a fifth conv; The output end of the first Ghost Bottleneck is connected to the input end of the second Ghost Bottleneck and the input end of the first concat respectively; the output end of the second Ghost Bottleneck is connected to the input end of the SPP and the input end of the second concat respectively; the output end of the SPP is connected to the input end of the fourth concat and the input end of the second upsample respectively; The output end of the second upsample is connected to the input end of the second concat, the output end of the second concat is connected to the input end of the fourth Ghost Bottleneck, the output end of the fourth Ghost Bottleneck is connected to the input end of the first upsample and the input end of the third concat respectively, the output end of the first upsample is connected to the input end of the first concat, the output end of the first concat is connected to the input end of the third Ghost Bottleneck Then, the output end of the third Ghost Bottleneck is connected to the input end of the first conv and the input end of the third conv respectively; The output end of the first conv is connected to the input end of the third concat, the output end of the third concat is connected to the input end of the fifth Ghost Bottleneck, the output end of the fifth Ghost Bottleneck is connected to the input end of the second conv and the input end of the fourth conv respectively, the output end of the second conv is connected to the input end of the fourth concat, and the output end of the fourth concat is connected to the input end of the sixth Ghost Bottleneck; The output end of the sixth Ghost Bottleneck is connected to the input end of the fifth conv.

3. The fall detection method according to claim 1, wherein: If the second judgment result is yes, after the step of confirming that the target in the image to be detected at the current moment has fallen, the method further includes: Sound an alarm.

4. A fall detection system, characterized in that: include: An acquisition module is used to obtain the image to be detected at the current moment; A first judgment module is used to process the image to be detected at the current moment using a lightweight YOLOv5s network to determine whether the target in the image to be detected at the current moment has fallen, and obtain a first judgment result; A second judgment module is configured to, if the first judgment result is yes, use OpenPose to process the image to be detected at the current moment, determine whether the target in the image to be detected at the current moment has fallen, and obtain a second judgment result; an alarm module, configured to confirm that the target in the image to be detected at the current moment has fallen down if the second judgment result is yes; A first returning module, configured to update the current moment and return to the step of obtaining the image to be detected at the current moment if the second judgment result is no; The second returning module is used to update the current moment and return to the step of obtaining the image to be detected at the current moment if the first judgment result is no.

5. The fall detection system according to claim 4, characterized in that The lightweight YOLOv5s network includes: SPP, a first Ghost Bottleneck, a second Ghost Bottleneck, a third Ghost Bottleneck, a fourth Ghost Bottleneck, a fifth Ghost Bottleneck, a sixth Ghost Bottleneck, a first upsample, a second upsample, a first concat, a second concat, a third concat, a fourth concat, a first conv, a second conv, a third conv, a fourth conv and a fifth conv; The output end of the first Ghost Bottleneck is connected to the input end of the second Ghost Bottleneck and the input end of the first concat respectively; the output end of the second Ghost Bottleneck is connected to the input end of the SPP and the input end of the second concat respectively; the output end of the SPP is connected to the input end of the fourth concat and the input end of the second upsample respectively; The output end of the second upsample is connected to the input end of the second concat, the output end of the second concat is connected to the input end of the fourth Ghost Bottleneck, the output end of the fourth Ghost Bottleneck is connected to the input end of the first upsample and the input end of the third concat respectively, the output end of the first upsample is connected to the input end of the first concat, the output end of the first concat is connected to the input end of the third Ghost Bottleneck, and the output end of the third Ghost Bottleneck is connected to the input end of the first conv and the input end of the third conv respectively; The output end of the first conv is connected to the input end of the third concat, the output end of the third concat is connected to the input end of the fifth Ghost Bottleneck, the output end of the fifth Ghost Bottleneck is connected to the input end of the second conv and the input end of the fourth conv respectively, the output end of the second conv is connected to the input end of the fourth concat, and the output end of the fourth concat is connected to the input end of the sixth Ghost Bottleneck; The output end of the sixth Ghost Bottleneck is connected to the input end of the fifth conv.

6. The fall detection system according to claim 4, wherein: Also includes: An alarm module is used to issue an alarm.

7. An electronic device, characterized in that: include: A memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the fall detection method according to any one of claims 1 to 3.

8. A computer-readable storage medium, characterized in that It stores a computer program, which, when executed by a processor, implements the fall detection method according to any one of claims 1 to 3.

Images