Intelligent fall detection method, device, equipment and medium

Abstract

The present invention provides an intelligent fall detection method, device, equipment and medium, which relates to the field of intelligent detection technology. The method includes: collecting video stream information in the environment and preprocessing it to obtain processed picture frames; extracting human motion paradigms in the processed picture frames to obtain human motion images; using a lightweight FDAttention network to perform behavioral analysis on the human motion images to determine whether the human body is in a fall state. By adopting advanced visual technology, the present invention can monitor the activity status of the elderly in real time and accurately, reduce the impact of scene complexity on subsequent data processing, reduce computing costs, and solve the problems of high misjudgment or missed judgment rates and high computing costs in existing intelligent fall detection technologies.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection technology, and in particular to an intelligent fall detection method, device, equipment and medium. Background Art

[0002] Population aging has become a normal problem. With the gradual deterioration of the elderly's physical functions and the aggravation of diseases, the risk of falling in the home environment or public areas has increased significantly. If the elderly do not receive timely treatment after falling, it may lead to further deterioration of their own diseases and even put their lives in danger. At present, manual real-time monitoring of the elderly is the most common means of monitoring, but it is time-consuming and labor-intensive.

[0003] Common fall detection devices and methods include three types: environmental sensor-based, wearable-based, and camera-based. Environmental sensor-based devices and methods have the problems of difficult layout and high false alarm rate. The deployment of environmental sensors requires careful planning and layout to ensure coverage of the area, which is more difficult to implement in some special home scenarios; and environmental sensors are usually more sensitive to other moving objects, which often leads to false alarms. Wearable devices and methods require the elderly to wear the device in real time, which is very inconvenient for the elderly. At the same time, wearable devices usually require regular charging and maintenance, and have a low usage rate.

[0004] Camera-based devices and methods usually first capture video streams, and then use deep learning methods to perform posture analysis, motion analysis, and shape analysis on the human body in the video to determine whether the elderly have fallen. Compared with the first two methods, the camera-based method is simple to deploy and non-invasive. However, it still has some challenges:

[0005] (1) Complex scenes: In complex scenes, such as those with large lighting changes, cluttered backgrounds, and occlusions, fall detection methods based on video streams may face great challenges. These challenges may make it difficult for the algorithm to accurately identify and analyze fall events, because visual features may be interfered with by the environment and cause misjudgment or omission;

[0006] (2) High computational overhead: Some detection methods usually require a lot of image processing and analysis, which places high demands on computing resources. Summary of the invention

[0007] In view of this, the embodiments of the present application provide an intelligent fall detection method, device, equipment and medium, aiming to effectively solve the problems of high misjudgment or missed judgment rate and high computational overhead in existing detection technologies. The embodiments of the present invention can monitor the activity status of the elderly in real time and accurately by adopting advanced visual technology, thereby reducing the impact of scene complexity on subsequent data processing and reducing computational costs.

[0008] The embodiment of the present application provides the following technical solution: an intelligent fall detection method, comprising:

[0009] Collect video stream information in the environment and perform preprocessing to obtain processed picture frames;

[0010] Extracting the human body motion paradigm in the processed picture frame to obtain a human body motion image;

[0011] A lightweight FDAttention is used to analyze the human body behavior on the human motion image to determine whether the human body is in a falling state.

[0012] According to an embodiment of the present application, collecting video stream information in an environment and preprocessing it includes:

[0013] Use high-sensitivity image sensors to capture original images in low-light environments;

[0014] Performing Fourier transform on the collected original image to convert the image from the spatial domain to the frequency domain;

[0015] Setting a learnable weight in the frequency domain, wherein the learnable weight is used to filter low-frequency noise and enhance specific high-frequency components to improve image contrast and details;

[0016] The processed frequency domain image is inverse Fourier transformed and converted back to the spatial domain to obtain the enhanced image.

[0017] According to an embodiment of the present application, extracting the human body motion paradigm in the processed picture frame to obtain a human body motion image includes:

[0018] Determine the approximate position of the human body: perform background subtraction processing on the picture frame to separate the moving human body from the environmental background; binarize the moving human body to obtain a binary image; use a morphological closing operation to complete the human body in the binary image, and use the completed maximum envelope space as the human body boundary box;

[0019] Precise human body positioning: The SAM model is used to finely segment the human body in the human body boundary box to generate a segmentation mask for precise human body positioning;

[0020] Environmental factor determination: Segment all objects in the picture frame using the SAM model, select an object mask whose intersection-over-union ratio with the segmentation mask accurately positioned with the human body is greater than 0.5, and input the object mask into the BLIP2 image-text annotation model to perform text annotation on all objects;

[0021] Action paradigm extraction: When the text of the object is an environmental object that affects human movement, it is retained; otherwise, it is removed, and finally a human motion image containing an action paradigm is obtained.

[0022] According to an embodiment of the present application, based on the human motion image, lightweight FDAttention is used to perform behavior analysis on the human body to determine whether the human body is in a falling state, including:

[0023] Acquire N consecutive frames of the human motion image, and use lightweight FDAttention to perform behavior analysis on the human motion image to determine whether the human body is in a falling state; wherein N is a positive integer, and N≥10.

[0024] According to an embodiment of the present application, a lightweight FDAttention is used to judge whether the human body is in a falling state on the human body motion image, including:

[0025] According to the human body motion image, extracting a single-frame spatial feature of a fall from the single-frame human body motion image;

[0026] According to the single-frame spatial features, obtaining the temporal features of the continuous fall events;

[0027] Obtaining a fall confidence level according to the temporal characteristics of the continuous fall events;

[0028] The fall confidence is compared with a set confidence threshold. If the fall confidence is greater than the set confidence threshold, it is determined that the human body is in a fall state.

[0029] According to an embodiment of the present application, lightweight FDAttention is used to analyze the position data of the key points of the human skeleton to determine whether the human body is in a falling state, and further includes:

[0030] The set confidence threshold includes a first confidence threshold and a second confidence threshold, and the fall confidence is compared with the set confidence threshold;

[0031] If the fall confidence is greater than the first confidence threshold and less than the second confidence threshold, a level one alarm is triggered;

[0032] If the fall confidence is greater than the second confidence threshold, a level 2 alarm is triggered.

[0033] The present application also provides an intelligent fall detection device, comprising:

[0034] An image acquisition module is used to collect video stream information in the environment and perform preprocessing to obtain processed picture frames;

[0035] An image acquisition module is used to collect video stream information in the environment and perform preprocessing to obtain processed picture frames;

[0036] The fall detection module is used to extract the human body movement paradigm in the processed image frame to obtain a human body motion image; and use a lightweight FDAttention network to perform human behavior analysis on the human body motion image in the human body image to determine whether the human body is in a fall state.

[0037] The present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned intelligent fall detection method when executing the computer program.

[0038] The present application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for executing the above-mentioned intelligent fall detection method.

[0039] Compared with the prior art, at least one of the above-mentioned technical solutions adopted in the embodiments of this specification can achieve at least the following beneficial effects: the embodiments of the present invention propose an FDAttention network to judge whether the elderly person has fallen based on the elderly person's motion images, thereby solving the problems of high misjudgment or missed judgment rates and high computational overhead in existing detection technologies, thereby real-time and accurate monitoring of the elderly person's activity status, and quickly issuing an alarm when a fall event is detected; it can effectively reduce the impact of the complexity of the scene on subsequent data processing, and also reduce the computational cost overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 is a flow chart of an intelligent fall detection method according to an embodiment of the present invention;

[0042] Figure 2 is a schematic structural diagram of an intelligent fall detection device according to a first embodiment of the present invention;

[0043] Figure 3 is a schematic structural diagram of an intelligent fall detection device according to a second embodiment of the present invention;

[0044] Figure 4 is a structural schematic diagram of an intelligent fall detection device according to a second embodiment of the present invention;

[0045] Figure 5is a schematic diagram of a lightweight FDAttention structure in an embodiment of the present invention;

[0046] Figure 6 It is a structural schematic diagram of the computer device of the present invention. DETAILED DESCRIPTION

[0047] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0048] The following describes the implementation methods of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the absence of conflict, the following embodiments and the features in the embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work belong to the scope of protection of the present application.

[0049] like Figure 1 As shown, an embodiment of the present invention provides an intelligent fall detection method, including:

[0050] S101. Collecting video stream information in the environment and preprocessing it to obtain processed picture frames;

[0051] S102. extracting the human motion paradigm in the processed picture frame to obtain a human motion image;

[0052] S103. According to the human body motion image, use the FDAttention network to perform behavior analysis on the human body to determine whether the human body is in a falling state.

[0053] In this embodiment, image acquisition and preprocessing are performed in a low-light environment through the following steps: Image acquisition: using a high-sensitivity image sensor to acquire the original image, Fourier transform: performing Fourier transform on the acquired original image to convert the image from the spatial domain to the frequency domain; High-frequency enhancement: setting a learnable weight in the frequency domain to filter low-frequency noise and enhance specific high-frequency components, thereby improving the contrast and details of the image; Inverse Fourier transform: performing an inverse Fourier transform on the processed frequency domain image and converting it back to the spatial domain to obtain an enhanced image. In the preprocessing process of this embodiment, Fourier transform processing is used to enable the preprocessing to adapt to low-light environments; high-frequency enhancement processing is performed on the video stream information to improve the quality of the image in low-light conditions.

[0054] The human motion paradigm can be viewed as a two-tuple (human motion, objects involved in the motion), which is specifically represented by an image that contains the human motion position and all objects involved in the environment.

[0055] In order to improve the accuracy of human motion paradigm extraction, the present invention proposes a multi-stage method for extracting human motion paradigm: In this embodiment, the motion paradigm in the processed image frame is extracted to obtain a human motion image, which specifically includes: (1) determining the approximate position of the human body; (2) accurately locating the human body; (3) determining environmental factors; and (4) extracting the motion paradigm.

[0056] In specific implementation, the step of determining the approximate position of the human body is specifically as follows:

[0057] (1) Separate the moving human body from the environmental background through KNN background subtraction;

[0058] (2) Binarizing the moving human body using the Otsu method to obtain a binary image;

[0059] (3) For the binary image, this embodiment uses a morphological closing operation to complete the shape, and regards the completed maximum envelope space as the body boundary box of the elderly.

[0060] In specific implementation, the human body precise positioning adopts the SAM (Segment Anything Model) model to use the human body boundary box determined in the first step as an input prompt, and then finely segment the human body to generate a segmentation mask for precise human body positioning.

[0061] In specific implementation, the environmental factor is determined by using the SAM model to segment all objects in the picture frame, selecting an object mask whose intersection with the segmentation mask that accurately locates the human body is greater than 0.5, and inputting the object mask into the BLIP2 image and text annotation model to perform text annotation on all objects.

[0062] In the specific implementation, when the text of the object is an environmental object that affects human movements, it is retained, otherwise, irrelevant objects are removed. Finally, the "Crop&Mask" method is used to obtain an image containing an action paradigm. Among them, the environmental objects that affect human movements are "bed", "stool" and other environmental objects that affect human movements.

[0063] This embodiment proposes an action paradigm extraction under environmental considerations, which takes into account both the actions of the human body and the environment in which the human body is located at the time.

[0064] In this embodiment, based on the motion image of the elderly, a lightweight FDAttention network is used to perform behavior analysis on the elderly to determine whether the human body is in a falling state, including: obtaining the human body motion image in N consecutive frames of human body images, and using a lightweight FDAttention network to analyze the human body motion image to determine whether the human body is in a falling state; wherein N is a positive integer, and N≥10.

[0065] The core function of the lightweight spatiotemporal linear network is to receive N consecutive frames of human skeleton key point data and output a signal of whether a fall has occurred based on these data. Usually, the system sets N to 10 frames. The lightweight FDAttention network is used to analyze the position data of the human skeleton key points to determine whether the human body is in a fall state, specifically including:

[0066] According to the human motion image, a single-frame spatial feature of a fall is extracted from the single-frame human motion image; according to the single-frame spatial feature, a temporal feature of continuous fall events is obtained; according to the temporal feature of the continuous fall events, a fall confidence is obtained; the fall confidence is compared with a set confidence threshold, and if the fall confidence is greater than the set confidence threshold, it is determined that the human body is in a fall state.

[0067] In the specific implementation, the spatial features of falling are extracted from a single-frame human motion image through the spatial feature extraction module. The spatial feature extraction module is built based on ResNet18, using the convolutional layer module of ResNet18 and adding an average pooling layer, and the weights are shared between the spatial feature modules. The human motion image is taken as input, and then a 256-dimensional vector is output, which is used to represent the single-frame spatial features of falling. The temporal feature extraction module is used to extract the temporal features of continuous falling events; the temporal feature extraction module is composed of 6 layers of Transformer Encoder layers, which include position embedding, self-attention mechanism layer, and fully connected layer. The temporal feature extraction module receives the N×256-dimensional vector output by the spatial feature extraction module, and outputs a vector with the same dimension as the spatial feature, which represents the temporal features of continuous falling events. This design enables the system to capture the evolution of falling events and further improves the accuracy of detection.

[0068] In this embodiment, the lightweight FDAttention is used to analyze the human motion image to determine whether the human body is in a falling state, and further includes:

[0069] The set confidence threshold includes a first confidence threshold and a second confidence threshold, and the fall confidence is compared with the set confidence threshold;

[0070] If the fall confidence is greater than the first confidence threshold and less than the second confidence threshold, a level one alarm is triggered;

[0071] If the fall confidence is greater than the second confidence threshold, a level 2 alarm is triggered.

[0072] In the specific implementation, the fall detection module is used for behavior analysis. The module consists of a linear layer, receives the 256-dimensional vector output by the time feature extraction module, and outputs a fall confidence. When the confidence is greater than 0.5 (i.e., the first confidence threshold), the system reminds the elderly’s contacts to check the elderly’s status through the WIFI module (i.e., triggering the first level alarm); when the confidence is greater than 0.7 (i.e., the second confidence threshold), the system sounds an alarm through the buzzer as a warning (i.e., triggering the second level alarm).

[0073] This embodiment uses lightweight FDAttention to perform behavioral analysis on the elderly to determine whether the human body is in a falling state, and can simultaneously determine whether a fall occurs on multiple image frames.

[0074] like Figure 2-Figure 5 As shown, the present application also provides an intelligent fall detection device 100, comprising:

[0075] The image acquisition module 110 is used to collect video stream information in the environment and perform preprocessing to obtain processed picture frames;

[0076] The fall detection module 120 is used to extract the human body motion paradigm in the processed image frame to obtain a human body motion image; based on the human body motion image, a lightweight FDAttention is used to perform human body behavior analysis to determine whether the human body is in a fall state.

[0077] In specific implementation, this embodiment further includes an alarm module 130, wherein the alarm module 130 includes a remote alarm module 131 and a buzzer alarm module 132. The remote alarm module 131, specifically a WIFI module, can send the fall information to the mobile phone of the elderly's relatives in real time; the buzzer alarm module 132, when detecting the elderly's fall, will issue a clear and obvious sound prompt to attract the attention of nearby people to provide emergency assistance.

[0078] The working process of the intelligent fall detection device 100 of the present invention is as follows:

[0079] (1) The image acquisition module 110 first acquires the environment video stream;

[0080] (2) The fall detection module 120 determines whether the elderly person falls based on the video stream collected by the image acquisition module 110.

[0081] (3) If the fall detection module 120 determines that the elderly person has fallen, the alarm module 130 will work. The remote alarm module 131 will send a text message to the mobile phone of the elderly person's relatives, and the buzzer alarm module 132 will issue a clear and obvious sound prompt to attract the attention of nearby people to provide emergency assistance.

[0082] The fall detection module 120 in the embodiment of the present invention is specifically deployed in an embedded device to detect whether an elderly person falls based on the video stream information; it can not only meet the computing performance requirements, but also help reduce the power consumption of the system, and is suitable for long-term monitoring tasks. The fall detection module 120 specifically includes an image preprocessing module 201, a motion paradigm detection module 202, and a behavior analysis module 203. The overall process of the intelligent detection method in the embodiment of the present invention is:

[0083] (1) Using the image preprocessing module 201 to improve the image quality, so as to reduce the impact of dark images collected in complex environments;

[0084] (2) The motion pattern detection module 202 receives the image quality improved in step (1) as input and outputs the motion image of the elderly in the image;

[0085] (3) The behavior analysis module 203 performs behavior recognition on the elderly based on the elderly movement image in step (2), and finally outputs the result of whether the elderly has fallen.

[0086] In specific implementation, the image preprocessing module 201 mainly includes using a high-sensitivity image sensor to collect original images in a low-light environment; performing Fourier transform on the collected original image to convert the image from the spatial domain to the frequency domain; setting a learnable weight in the frequency domain, and the learnable weight is used to filter low-frequency noise and enhance specific high-frequency components to improve image contrast and details; performing an inverse Fourier transform on the processed frequency domain image and converting it back to the spatial domain to obtain an enhanced image.

[0087] In specific implementation, the motion paradigm detection module 202 mainly includes:

[0088] Determine the approximate position of the human body: perform background subtraction processing on the picture frame to separate the moving human body from the environmental background; binarize the moving human body to obtain a binary image; use a morphological closing operation to complete the human body in the binary image, and use the completed maximum envelope space as the human body boundary box;

[0089] Precise human body positioning: The SAM model is used to finely segment the human body in the human body boundary box to generate a segmentation mask for precise human body positioning;

[0090] Environmental factor determination: Segment all objects in the picture frame using the SAM model, select an object mask whose intersection-over-union ratio with the segmentation mask accurately positioned with the human body is greater than 0.5, and input the object mask into the BLIP2 image-text annotation model to perform text annotation on all objects;

[0091] Action paradigm extraction: When the text of the object is an environmental object that affects human movement, it is retained; otherwise, it is removed, and finally a human motion image containing an action paradigm is obtained.

[0092] In the specific implementation, the behavior analysis module 203 is constructed based on FDAttention. Furthermore, in order to reduce the computing resource overhead, the embodiment of the present invention implements 8-bit quantization operation on the original FDAttention weight. Specifically, the behavior analysis module 203 mainly includes the following steps:

[0093] (1) The motion image of the elderly is detected by the motion pattern detection module 202, and the motion image containing the elderly is cropped from the original image; (2) The motion image is subjected to a behavior analysis of the human body using a lightweight FDAttention to determine whether the human body is in a falling state.

[0094] The behavior analysis module 203 is further analyzed based on the motion image extracted by the motion paradigm detection module 202. The embodiment of the present invention proposes a lightweight FDAttention network to analyze the motion image extracted by the motion paradigm extraction module 202. The lightweight FDAttention receives N consecutive frames of human skeleton key points and outputs a signal of whether the person falls. N is usually set to 10. FDAttention can be divided into the following three modules: a spatial feature extraction module, a temporal feature extraction module, and a fall detection module; the spatial feature extraction module is used to extract a single-frame spatial feature of a fall from a single-frame human motion image according to the human motion image; the temporal feature extraction module is used to obtain the temporal features of continuous fall events according to the single-frame spatial features; the fall detection module is used to obtain the fall confidence according to the temporal features of the continuous fall events; the fall confidence is compared with the set confidence threshold, and if the fall confidence is greater than the set confidence threshold, it is judged that the human body is in a fall state.

[0095] In one embodiment, a computer device is provided, such as Figure 6 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned intelligent fall detection method is implemented.

[0096] Specifically, the computer device may be a computer terminal, a server, a microprocessor or a similar computing device.

[0097] In this embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program for executing the above-mentioned intelligent fall detection method.

[0098] Specifically, computer-readable storage media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer-readable storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0099] Obviously, those skilled in the art should understand that the modules or steps of the above-mentioned embodiments of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the embodiments of the present invention are not limited to any specific combination of hardware and software.

[0100] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. An intelligent fall detection method, characterized in that: include: Collect video stream information in the environment and perform preprocessing to obtain processed picture frames; Extracting the human body motion paradigm in the processed picture frame to obtain a human body motion image; A lightweight FDAttention is used to analyze the human body's behavior on the human body motion image to determine whether the human body is in a falling state; Extracting the human body motion paradigm in the processed picture frame to obtain a human body motion image, including: Determine the approximate position of the human body: perform background subtraction processing on the picture frame to separate the moving human body from the environmental background; binarize the moving human body to obtain a binary image; use a morphological closing operation to complete the human body in the binary image, and use the completed maximum envelope space as the human body boundary box; Precise human body positioning: The SAM model is used to finely segment the human body in the human body boundary box to generate a segmentation mask for precise human body positioning; Environmental factor determination: Segment all objects in the picture frame using the SAM model, select an object mask whose intersection-over-union ratio with the segmentation mask accurately positioned with the human body is greater than 0.5, and input the object mask into the BLIP2 image-text annotation model to perform text annotation on all objects; Action paradigm extraction: When the text of the object is an environmental object that affects human movement, it is retained; otherwise, it is removed, and finally a human motion image containing an action paradigm is obtained.

2. The intelligent fall detection method according to claim 1, characterized in that: Collect video stream information in the environment and perform preprocessing, including: Use high-sensitivity image sensors to capture original images in low-light environments; Performing Fourier transform on the collected original image to convert the image from the spatial domain to the frequency domain; Setting a learnable weight in the frequency domain, wherein the learnable weight is used to filter low-frequency noise and enhance specific high-frequency components to improve image contrast and details; The processed frequency domain image is inverse Fourier transformed and converted back to the spatial domain to obtain the enhanced image.

3. The intelligent fall detection method according to claim 1, characterized in that: According to the human motion image, a lightweight FDAttention is used to analyze the human behavior to determine whether the human body is in a falling state, including: Acquire N consecutive frames of the human motion image, and use lightweight FDAttention to perform behavior analysis on the human motion image to determine whether the human body is in a falling state; wherein N is a positive integer, and N≥10.

4. The intelligent fall detection method according to claim 3, characterized in that: A lightweight FDAttention network is used to analyze the human motion images to determine whether the human body is in a falling state, including: According to the human body motion image, extracting a single-frame spatial feature of a fall from the single-frame human body motion image; According to the single-frame spatial features, obtaining the temporal features of the continuous fall events; Obtaining a fall confidence level according to the temporal characteristics of the continuous fall events; The fall confidence is compared with a set confidence threshold. If the fall confidence is greater than the set confidence threshold, it is determined that the human body is in a fall state.

5. The intelligent fall detection method according to claim 4, characterized in that: The lightweight FDAttention network is used to analyze the human motion image to determine whether the human body is in a falling state, and further includes: The set confidence threshold includes a first confidence threshold and a second confidence threshold, and the fall confidence is compared with the set confidence threshold; If the fall confidence is greater than the first confidence threshold and less than the second confidence threshold, a level one alarm is triggered; If the fall confidence is greater than the second confidence threshold, a level 2 alarm is triggered.

6. An intelligent fall detection device, using the intelligent fall detection method according to any one of claims 1 to 5, characterized in that: include: An image acquisition module is used to collect video stream information in the environment and perform preprocessing to obtain processed picture frames; The fall detection module is used to extract the human body movement paradigm in the processed image frame to obtain a human body movement image; a lightweight FDAttention network is used to perform human behavior analysis on the human body movement image to determine whether the human body is in a fall state.

7. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the intelligent fall detection method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program for executing the intelligent fall detection method according to any one of claims 1 to 5.

Images