A fall detection method, device, electronic device and storage medium

Abstract

The present invention provides a fall detection method, device, electronic device and storage medium. The method includes: performing human skeleton key point detection on a human body video image to obtain initial human skeleton key point data; inputting the initial human skeleton key point data into a preset skeleton key point correction model to obtain corrected human skeleton key point data; extracting human skeleton target key point data from the corrected human skeleton key point data that meets the data input conditions of a preset fall discrimination model; inputting the human skeleton target key point data into the preset fall discrimination model to obtain a fall discrimination result; wherein, the preset skeleton key point correction model is trained according to first human skeleton key point sample data carrying preset human skeleton key point labels; and wherein, the preset fall discrimination model is trained according to second human skeleton key point sample data carrying fall result labels.

Description

Technical Field

[0001] The present invention relates to the technical field of information processing, and particularly to a fall detection method, device, electronic device, and storage medium. Background Art

[0002] For the research on fall detection, the mainstream methods mainly include three types: fall detection based on wearable sensors, fall detection based on environmental sensors, and fall detection methods based on video images. Among them, the fall detection method based on video images does not require the human body to wear a device and is not easily affected by the environment, and has been more widely used in practical applications in recent years.

[0003] In the prior art, the fall detection method based on video images mainly relies on manually extracted fall features and uses a discriminant classifier for fall detection. The model is simple but has a relatively low accuracy. In the prior art, another commonly used method is to use a depth camera such as Kinect to collect video data, obtain the three-dimensional coordinate information of the human body, and then combine the changes in the angles and distances of the human body bone points and a machine learning classifier to perform human fall detection and judgment. Such methods have high requirements for the camera, increase the hardware cost, and have a relatively high algorithm complexity.

[0004] Therefore, how to better detect falls has become an urgent problem to be solved in the industry. Summary of the Invention

[0005] The present invention provides a fall detection method, device, electronic device, and storage medium to solve the problem in the prior art that fall detection cannot be conveniently and efficiently implemented.

[0006] The present invention provides a fall detection method, including:

[0007] Performing human bone key point detection on a human body video image to obtain initial human bone key point data;

[0008] Inputting the initial human bone key point data into a preset bone key point correction model to obtain corrected human bone key point data;

[0009] Extracting human bone target key point data from the corrected human bone key point data that meets the data input conditions of a preset fall discrimination model;

[0010] Inputting the human bone target key point data into a preset fall discrimination model to obtain a fall discrimination result;

[0011] Wherein, the preset bone key point correction model is trained according to first human bone key point sample data carrying preset human bone key point labels;

[0012] Among them, the preset fall discrimination model is trained according to the second human skeletal key point sample data with fall result labels.

[0013] According to a fall detection method provided by the present invention, the corrected human skeletal key point data that meets the data input conditions of the preset fall discrimination model is specifically:

[0014] The corrected human skeletal key point data includes both head key point data and at least any one complete leg key point data.

[0015] According to a fall detection method provided by the present invention, the steps of extracting human skeletal target key point data from the corrected human skeletal key point data that meets the data input conditions of the preset fall discrimination model specifically include:

[0016] Screen out the head key point data and the complete leg key point data from the corrected human skeletal key point data that meets the data input conditions of the preset fall discrimination model;

[0017] Use the head key point data and any one complete leg key point data as the human skeletal target key point data.

[0018] According to a fall detection method provided by the present invention, before the step of inputting the initial human skeletal key point data into the preset skeletal key point correction model, the method further includes:

[0019] Obtain a human video sample set, extract the openpose features of the first frame image in the human video sample set through the openpose algorithm to obtain an initial hidden state;

[0020] Obtain human video sample pictures in the human video sample set at a preset interval of frames;

[0021] Perform human skeletal key point detection on the human video sample pictures to obtain a plurality of first human skeletal key point sample data;

[0022] Use the combination of each first human skeletal key point sample data and the preset human skeletal key point label as a training sample to obtain a plurality of training samples;

[0023] Train the gated recurrent unit GRU recurrent neural network with a plurality of training samples and the initial hidden state. When the first preset condition is met, the training is completed to obtain a preset skeletal key point correction model.

[0024] According to a fall detection method provided by the present invention, before the step of inputting the human skeletal target key point data into the preset fall discrimination model to obtain a fall discrimination result, the method further includes:

[0025] Obtain a plurality of second human skeletal key point sample data, perform scale transformation on the plurality of second human skeletal key point sample data to obtain second human skeletal key point sample data of multiple different scales;

[0026] Extract human skeletal target key point sample data according to the second human skeletal key point sample data of each different scale;

[0027] Take the combination of each of the human skeletal target key point sample data and the fall result label as a training sample to obtain a plurality of training samples;

[0028] Train a preset feature extraction network according to the plurality of training samples, and when the second preset condition is satisfied, complete the training to obtain a preset fall discrimination model.

[0029] According to a fall detection method provided by the present invention, the step of performing human skeletal key point detection on a human body video image to obtain initial human skeletal key point data specifically includes:

[0030] Extract the openpose feature of the human body video image through the openpose algorithm;

[0031] Perform human skeletal key point detection according to the openpose feature to obtain initial human skeletal key point data.

[0032] The present invention also provides a fall detection device, including:

[0033] An identification module for performing human skeletal key point detection on a human body video image to obtain initial human skeletal key point data;

[0034] A correction module for inputting the initial human skeletal key point data into a preset skeletal key point correction model to obtain corrected human skeletal key point data;

[0035] An extraction module for extracting human skeletal target key point data from the corrected human skeletal key point data that meets the data input conditions of the preset fall discrimination model;

[0036] A detection module for inputting the human skeletal target key point data into a preset fall discrimination model to obtain a fall discrimination result;

[0037] Among them, the preset skeletal key point correction model is trained according to the first human skeletal key point sample data carrying preset human skeletal key point labels;

[0038] Among them, the preset fall discrimination model is trained according to the second human skeletal key point sample data carrying fall result labels.

[0039] The extraction module is specifically configured to:

[0040] From the corrected human body bone key point data that meets the data input conditions of the preset fall discrimination model, screen out the head key point data and the complete leg key point data;

[0041] Use the head key point data and any complete leg key point data as the human body bone target key point data.

[0042] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of any one of the above-mentioned fall detection methods are implemented.

[0043] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the above-mentioned fall detection methods are implemented.

[0044] A fall detection method, device, electronic device, and storage medium provided by the present invention obtain initial human body bone key point data through human body video data collected by an ordinary camera and common human body bone key point detection. Then, the initial human body bone key point data is corrected by a preset bone key point correction model to reduce the misdetection rate of bone key points and ensure the accuracy of bone key point detection. Subsequently, through a fall detection model based on a deep learning classification model, the local neighborhood features and global features of bone key points at different scales are fused to further improve the accuracy and robustness of fall detection in actual application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0046] Figure 1 is a flowchart of the fall detection method provided by the present invention;

[0047] Figure 2 is a schematic diagram of the positions and serial numbers of human body bone key points provided by the present invention;

[0048] Figure 3 is a schematic diagram of the human body bone point correction module of the GRU unit provided by the present invention;

[0049] Figure 4The flowchart of fall detection in the actual scenario provided by the present invention;

[0050] Figure 5 The structural schematic diagram of the fall detection device provided by the present invention;

[0051] Figure 6 The entity structure schematic diagram of the electronic device provided by the present invention. Detailed implementation manners

[0052] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0053] Figure 1 It is the flowchart of the fall detection method provided by the present invention. As Figure 1 shown, it includes:

[0054] Step S1: Detect the key points of the human body skeleton in the human body video image to obtain the initial human body skeleton key point data;

[0055] The present invention can perform a preliminary detection of the key points of the human body skeleton based on openpose, or can also implement the detection of the key points of the human body skeleton based on other common algorithms. The present invention does not limit this.

[0056] In the initial human body skeleton key point data described in the present invention, it contains the key points of the main human body skeletons. Figure 2 It is the schematic diagram of the position and serial number of the key points of the human body skeleton provided by the present invention. As Figure 2 shown, when performing fall detection, the key points required are mainly the head key points (serial number 1) and the complete leg key points (the set of serial numbers 8, 9, 10 or the set of serial numbers 11, 12, 13).

[0057] In complex scenarios or when the distance between people is relatively close, there is a certain probability of false detection in the commonly used algorithms for detecting the key points of the human body skeleton, that is, the background is misdetected as a certain bone point of the human body or some bone points of adjacent people are misjudged with each other, which will greatly affect the accuracy of the fall detection result.

[0058] Step S2: Input the initial human body skeleton key point data into a preset skeleton key point correction model to obtain the corrected human body skeleton key point data;

[0059] Since there is a certain probability of false detection in the human skeleton key point data obtained only by common human skeleton key point detection algorithms, and in order to avoid the impact of false key point detection on fall detection, in the present invention, based on the GRU unit of the recurrent neural network, the human skeleton key points detected by the common human skeleton key point detection algorithm are corrected twice, which can effectively correct the reliability of the human skeleton key points, thereby further ensuring the accuracy and reliability of fall detection.

[0060] Step S3: Extract the human skeleton target key point data from the corrected human skeleton key point data that meets the preset fall discrimination model data input conditions;

[0061] The preset fall discrimination model data input conditions in the present invention are used to determine whether the corrected human skeleton key point data contains the head key point data and at least any complete leg key point data necessary for fall detection.

[0062] By setting the preset fall discrimination model data input conditions, the present invention filters out the data that does not meet the fall detection requirements, effectively improving the efficiency and speed of data processing.

[0063] At the same time, by extracting the human skeleton target key point data from the corrected human skeleton key point data, the present invention can effectively reduce the amount of data input into the preset fall discrimination model and improve the efficiency of data processing.

[0064] Step S4: Input the human skeleton target key point data into the preset fall discrimination model to obtain a fall discrimination result;

[0065] Among them, the preset skeleton key point correction model is trained according to the first human skeleton key point sample data carrying the preset human skeleton key point label;

[0066] Among them, the preset fall discrimination model is trained according to the second human skeleton key point sample data carrying the fall result label.

[0067] Specifically, through training, the preset fall discrimination model in the present invention fuses the local neighborhood features and global features of the skeleton key points at different scales, further improving the accuracy and robustness of fall detection in actual application scenarios.

[0068] The present invention collects human body video data through an ordinary camera, obtains initial human body bone key point data through common human body bone key point detection, and then corrects the initial human body bone key point data through a preset bone key point correction model, reducing the misdetection rate of bone key points and ensuring the accuracy of bone key point detection. A fall detection model based on a deep learning classification model is proposed, which fuses the local neighborhood features and global features of bone key points at different scales, further improving the accuracy and robustness of fall detection in actual application scenarios.

[0069] Based on any of the above embodiments, the corrected human body bone key point data that meets the data input conditions of the preset fall discrimination model is specifically:

[0070] The corrected human body bone key point data simultaneously includes head key point data and at least any complete leg key point data.

[0071] Based on any of the above embodiments, the steps of extracting human body bone target key point data from the corrected human body bone key point data that meets the data input conditions of the preset fall discrimination model specifically include:

[0072] Screen out the head key point data and complete leg key point data from the corrected human body bone key point data that meets the data input conditions of the preset fall discrimination model;

[0073] Use the head key point data and any complete leg key point data as the human body bone target key point data.

[0074] Specifically, in the fall judgment event: Since the upper body and legs of the human body are the core reference points, while the bone points such as the head, left and right hands are non-core reference points and are prone to misdetection caused by occlusion or arm swinging, etc.; in addition, if the upper body and a complete leg key point of the human body cannot be observed, the human eye cannot judge whether the human body is in a falling state. Therefore, the condition for the input data of the fall discrimination model in this solution is: It is necessary to ensure that the corrected human body bone key point data simultaneously includes head key point data and at least any complete leg key point data to achieve fall judgment.

[0075] Therefore, specifically, it is necessary to ensure that the corrected human body bone key point data needs to ensure the simultaneous presence of the head (number 1), right waist (number 8), right knee (number 9), and right foot head (number 10) or the simultaneous presence of the head (number 1), left waist (number 11), left knee (number 12), and left foot head (number 13) key points.

[0076] The present invention effectively improves the efficiency and speed of data processing by setting the data input conditions of the preset fall discrimination model, thereby screening out data that does not meet the fall detection requirements.

[0077] Based on any of the above embodiments, before the step of inputting the initial human skeleton key point data into the preset skeleton key point correction model, the method further includes:

[0078] Obtain a set of human video samples, extract the openpose features of the first frame image in the set of human video samples through the openpose algorithm to obtain an initial hidden state;

[0079] Obtain human video sample pictures in the set of human video samples at a preset interval of frames;

[0080] Perform human skeleton key point detection on the human video sample pictures to obtain a plurality of first human skeleton key point sample data;

[0081] Take the combination of each first human skeleton key point sample data and the preset human skeleton key point label as a training sample to obtain a plurality of training samples;

[0082] Use the plurality of training samples and the initial hidden state to train the gated recurrent unit GRU recurrent neural network. When the first preset condition is satisfied, the training is completed to obtain the preset skeleton key point correction model.

[0083] Specifically, the preset interval of frames described in the present invention can be set according to requirements. It obtains video frames from the human video at a preset interval to obtain human video sample pictures.

[0084] The implementation manner of performing human skeleton key point detection on the human video sample pictures in the present invention is also the manner of the prior art, usually implemented through openpose.

[0085] The preset human skeleton key point label described in the present invention is the complete human skeleton key point information obtained by manually annotating a plurality of first human skeleton key point sample data.

[0086] The first preset condition described in the present invention may refer to satisfying a preset number of training times, or the value of the loss function being less than a preset threshold.

[0087] The preset skeleton key point correction model in the present invention is composed of a plurality of GRU algorithm units. During the operation of each GRU algorithm unit, it will obtain the hidden state of the previous node of the current node to prevent data loss. For the first initial node, the openpose features of the first frame image in the set of human video samples extracted through the openpose algorithm are used as the initial hidden state.

[0088] Figure 3 Schematic diagram of the human skeleton point correction module of the GRU unit provided by the present invention, as Figure 3As shown in the figure, for the trained GRU algorithm unit, the input is the GRU hidden node state information N frames before and the current frame human body bone key point data, and the output is the corrected current frame human body bone key point position information and the hidden state of the next node. The trained GRU unit module is used for the correction of the human body bone key points. Let the current frame number be ni, and the corresponding human body bone key point data predicted by openpose be X(ni); the frame number N frames before is ni - N, and the corresponding hidden state of the current node is H(ni - N). The corrected current frame human body bone key point data predicted by the GRU unit is Y(ni), and the hidden state of the next node is H(ni).

[0089] The present invention corrects the initial human body bone key point data through a preset bone key point correction model, reduces the misdetection rate of bone key points, ensures the accuracy of bone key point detection, and is beneficial to ensuring the accuracy of subsequent fall detection.

[0090] Based on any of the above embodiments, before the step of inputting the human body target key point data into a preset fall discrimination model to obtain a fall discrimination result, the method further includes:

[0091] Obtain a plurality of second human body bone key point sample data, perform scale transformation on the plurality of second human body bone key point sample data to obtain a plurality of second human body bone key point sample data with different scales;

[0092] Extract human body target key point sample data according to each second human body bone key point sample data with a different scale;

[0093] Take the combination of each human body target key point sample data and a fall result label as a training sample to obtain a plurality of training samples;

[0094] Train a preset feature extraction network according to the plurality of training samples. When the second preset condition is satisfied, the training is completed to obtain a preset fall discrimination model.

[0095] Specifically, since the commonly used deep learning fall discrimination model has a slightly weak ability to describe the local distribution features of adjacent limbs and is easily affected by scale changes. Therefore, this solution adds a multi-scale network structure to extract the multi-scale local features of some key points, improving the discrimination accuracy of the model in actual application scenarios.

[0096] In the embodiments of the present invention, networks such as ResNet and MobileNet can be selected as feature extraction networks. The second human skeleton key point sample data is transformed through three different scales to obtain input data of different scales. In each scale of input data, points numbered 8, 9, 11, and 12 and their corresponding connection points are selected as neighborhood points, and then the neighborhood points of different scales are input into the feature extraction network to extract the neighborhood local features corresponding to each point; all the neighborhood local features and the global features are combined through a fully connected layer for classification and calculation of the loss function.

[0097] The fall result label described in the present invention is a label annotated with fall status information.

[0098] The second preset condition in the present invention specifically refers to satisfying the preset number of training times or the value of the loss function being less than the preset threshold.

[0099] Figure 4 is the fall detection flowchart in the actual scenario provided by the present invention. As Figure 4 shown, in the actual scenario application, based on the openpose and GRU unit modules, the skeleton key points of all humans in the scenario graph are detected and corrected, and the single-person data that meets the input conditions of the fall discrimination model is sequentially input into the constructed fall discrimination model for fall detection judgment, and the single-person data that does not meet the input conditions is discarded for this fall detection discrimination.

[0100] Then, through the fall detection model based on the deep learning classification model, the present invention fuses the local neighborhood features and the global features of the skeleton key points at different scales, further improving the accuracy and robustness of fall detection in the actual application scenario.

[0101] Figure 5 is the structural schematic diagram of the fall detection device provided by the present invention. As Figure 5As shown in the figure, it includes: an identification module 510, a correction module 520, an extraction module 530, and a detection module 540; among them, the identification module 510 is used to detect human body bone key points in a human body video image to obtain initial human body bone key point data; among them, the correction module 520 is used to input the initial human body bone key point data into a preset bone key point correction model to obtain corrected human body bone key point data; among them, the extraction module 530 is used to extract human body bone target key point data from the corrected human body bone key point data that meets the data input conditions of a preset fall discrimination model; among them, the detection module 540 is used to input the human body bone target key point data into a preset fall discrimination model to obtain a fall discrimination result; among them, the preset bone key point correction model is trained according to the first human body bone key point sample data carrying preset human body bone key point labels; among them, the preset fall discrimination model is trained according to the second human body bone key point sample data carrying fall result labels.

[0102] The extraction module 530 is specifically used for:

[0103] Screen out the head key point data and the complete leg key point data from the corrected human body bone key point data that meets the data input conditions of the preset fall discrimination model;

[0104] Use the head key point data and any complete leg key point data as the human body bone target key point data.

[0105] The present invention collects human body video data through an ordinary camera, obtains initial human body bone key point data through common human body bone key point detection, and then corrects the initial human body bone key point data through a preset bone key point correction model to reduce the misdetection rate of bone key points and ensure the accuracy of bone key point detection. Then, through a fall detection model based on a deep learning classification model, the local neighborhood features and global features of bone key points at different scales are fused to further improve the accuracy and robustness of fall detection in actual application scenarios.

[0106] Figure 6 It is a schematic physical structure diagram of the electronic device provided by the present invention, as Figure 6As shown in the figure, the electronic device may include: a processor 610, a communications interface 620, a memory 630, and a communication bus 640. Among them, the processor 610, the communications interface 620, and the memory 630 complete communication with each other through the communication bus 640. The processor 610 may call the logical instructions in the memory 630 to execute a fall detection method, which includes: performing human skeleton key point detection on a human body video image to obtain initial human skeleton key point data; inputting the initial human skeleton key point data into a preset skeleton key point correction model to obtain corrected human skeleton key point data; extracting human skeleton target key point data from the corrected human skeleton key point data that meets the data input conditions of a preset fall discrimination model; inputting the human skeleton target key point data into the preset fall discrimination model to obtain a fall discrimination result; where the preset skeleton key point correction model is trained according to first human skeleton key point sample data carrying preset human skeleton key point labels; and where the preset fall discrimination model is trained according to second human skeleton key point sample data carrying fall result labels.

[0107] In addition, when the logical instructions in the above-mentioned memory 630 are implemented in the form of software functional units and sold or used as independent products, they may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, may be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0108] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the fall detection method provided by each of the above methods. The method includes: performing human skeleton key point detection on a human body video image to obtain initial human skeleton key point data; inputting the initial human skeleton key point data into a preset skeleton key point correction model to obtain corrected human skeleton key point data; extracting human skeleton target key point data from the corrected human skeleton key point data that meets the data input conditions of a preset fall discrimination model; inputting the human skeleton target key point data into the preset fall discrimination model to obtain a fall discrimination result; wherein, the preset skeleton key point correction model is trained according to first human skeleton key point sample data carrying preset human skeleton key point labels; and wherein, the preset fall discrimination model is trained according to second human skeleton key point sample data carrying fall result labels.

[0109] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the fall detection method provided by each of the above embodiments. The method includes: performing human skeleton key point detection on a human body video image to obtain initial human skeleton key point data; inputting the initial human skeleton key point data into a preset skeleton key point correction model to obtain corrected human skeleton key point data; extracting human skeleton target key point data from the corrected human skeleton key point data that meets the data input conditions of a preset fall discrimination model; inputting the human skeleton target key point data into the preset fall discrimination model to obtain a fall discrimination result; wherein, the preset skeleton key point correction model is trained according to first human skeleton key point sample data carrying preset human skeleton key point labels; and wherein, the preset fall discrimination model is trained according to second human skeleton key point sample data carrying fall result labels.

[0110] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative effort.

[0111] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fall detection method, characterized in that, Including: Performing human skeleton key point detection on a human body video image to obtain initial human skeleton key point data; Inputting the initial human skeleton key point data into a preset skeleton key point correction model to obtain corrected human skeleton key point data; Extracting human skeleton target key point data from the corrected human skeleton key point data that meets the data input conditions of a preset fall discrimination model; Inputting the human skeleton target key point data into a preset fall discrimination model to obtain a fall discrimination result; Wherein, the preset skeleton key point correction model is trained according to first human skeleton key point sample data carrying preset human skeleton key point labels; Wherein, the preset fall discrimination model is trained according to second human skeleton key point sample data carrying fall result labels; Wherein, before the step of inputting the initial human skeleton key point data into the preset skeleton key point correction model, the method further includes: Obtaining a human body video sample set, extracting openpose features of the first frame image in the human body video sample set through the openpose algorithm to obtain an initial hidden state; Obtaining human body video sample pictures in the human body video sample set at preset interval frames; Performing human skeleton key point detection on the human body video sample pictures to obtain a plurality of first human skeleton key point sample data; Taking the combination of each first human skeleton key point sample data and a preset human skeleton key point label as a training sample to obtain a plurality of training samples; Training a gated recurrent unit (GRU) recurrent neural network by using the plurality of training samples and the initial hidden state, and completing the training when a first preset condition is satisfied to obtain a preset skeleton key point correction model.

2. The fall detection method according to claim 1, wherein The corrected human skeleton key point data that meets the data input conditions of the preset fall discrimination model is specifically: The corrected human skeleton key point data simultaneously includes head key point data and at least any one complete leg key point data.

3. The fall detection method according to claim 2, wherein The step of extracting human skeleton target key point data from the corrected human skeleton key point data that meets the data input conditions of the preset fall discrimination model specifically includes: Screening out head key point data and complete leg key point data from the corrected human skeleton key point data that meets the data input conditions of the preset fall discrimination model; Taking the head key point data and any one complete leg key point data as human skeleton target key point data.

4. The fall detection method according to claim 1, characterized in that, Before the step of inputting the human skeleton target key point data into the preset fall discrimination model to obtain a fall discrimination result, the method further includes: Obtaining a plurality of second human skeleton key point sample data, performing scale transformation on the plurality of second human skeleton key point sample data to obtain a plurality of second human skeleton key point sample data with different scales; Extracting human skeleton target key point sample data according to each second human skeleton key point sample data with a different scale; Taking the combination of each human skeleton target key point sample data and a fall result label as a training sample to obtain a plurality of training samples; Train a preset feature extraction network according to multiple said training samples. When the second preset condition is satisfied, the training is completed to obtain a preset fall discrimination model.

5. The fall detection method according to claim 1, wherein The step of performing human skeleton key point detection on a human body video image to obtain initial human skeleton key point data specifically includes: Extract the openpose feature of the human body video image through the openpose algorithm; Perform human skeleton key point detection according to the openpose feature to obtain initial human skeleton key point data.

6. A fall detection device, characterized in that, It includes: An identification module for performing human skeleton key point detection on a human body video image to obtain initial human skeleton key point data; A correction module for inputting the initial human skeleton key point data into a preset skeleton key point correction model to obtain corrected human skeleton key point data; An extraction module for extracting human skeleton target key point data from the corrected human skeleton key point data that meets the data input conditions of the preset fall discrimination model; A detection module for inputting the human skeleton target key point data into a preset fall discrimination model to obtain a fall discrimination result; Among them, the preset skeleton key point correction model is trained according to the first human skeleton key point sample data carrying preset human skeleton key point labels; Among them, the preset fall discrimination model is trained according to the second human skeleton key point sample data carrying fall result labels; Among them, the device is further used for: Obtain a human body video sample set, extract the openpose feature of the first frame image in the human body video sample set through the openpose algorithm to obtain an initial hidden state; Obtain human body video sample pictures in the human body video sample set at a preset interval of frames; Perform human skeleton key point detection on the human body video sample pictures to obtain a plurality of first human skeleton key point sample data; Take the combination of each first human skeleton key point sample data and a preset human skeleton key point label as a training sample to obtain a plurality of training samples; Use the plurality of training samples and the initial hidden state to train a gated recurrent unit (GRU) recurrent neural network. When the first preset condition is satisfied, the training is completed to obtain a preset skeleton key point correction model.

7. The fall detection device according to claim 6, characterized in that, The extraction module specifically is used for: Screen out the head key point data and the complete leg key point data from the corrected human skeleton key point data that meets the data input conditions of the preset fall discrimination model; Take the head key point data and any complete leg key point data as the human skeleton target key point data.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the fall detection method according to any one of claims 1 to 5.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the fall detection method according to any one of claims 1 to 5.

Images