A human posture evaluation method, device and computer readable storage medium

Abstract

The present application relates to a kind of human posture evaluation method, device and computer readable storage medium, belong to image processing technical field;Each image in the video frame to be evaluated is input into target detection model, and the center point coordinate set of the image of the frame is obtained;If the center point coordinate set of the image of the frame is non-empty set, then the image of the frame is input into human skeleton extraction model, and the center point coordinate set of the human eyes of the image of the frame is obtained;The Euclidean distance between each center point coordinate of the image of the frame and each center point coordinate of the human eyes is calculated, and the minimum Euclidean distance obtained is used as the detection result of the image of the frame;The detection result of each image in the video frame to be evaluated is obtained, if the detection result of the preset number of image in the video frame to be evaluated is less than the preset threshold, and the preset number of image is continuous in time, then determine that there is a shooting behavior in the video frame to be evaluated.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method, apparatus and computer-readable storage medium for human posture assessment. Background Technology

[0002] Human posture assessment technology is an important research area in computer vision and robotics learning. This technology has broad application prospects in many fields such as security monitoring, human-computer interaction, virtual display and healthcare. For example, in the healthcare field, human posture assessment technology can be used to monitor and assess the posture and movement status of patients in real time, thereby assisting in rehabilitation treatment.

[0003] In certain specific scenarios, a human posture assessment method is needed to detect whether a user is taking photos. For example, in security- or privacy-sensitive areas such as research laboratories, photography is often prohibited to prevent the leakage of confidential materials. Detecting user photography through human posture assessment can promptly stop users and strengthen confidentiality measures. In applications such as social media, detecting user photography actions can automatically trigger filter effects, enhancing the user experience. Furthermore, in retail environments, merchants can use human posture assessment to identify products that are of high interest or popularity, enabling targeted marketing and product placement. However, existing human posture assessment methods are mostly used to evaluate whether a user's posture conforms to standards for a specific scenario, such as whether a badminton player's serve is standard, and cannot detect user photography actions.

[0004] In summary, a human posture assessment method is needed to detect user photo-taking behavior in specific scenarios. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the problem that the human posture assessment method in the prior art cannot detect the user's photo-taking behavior.

[0006] To address the aforementioned technical problems, this invention provides a method for human posture assessment, comprising:

[0007] The video frame to be evaluated is obtained, and each frame image in the video frame to be evaluated is input into the target detection model. The coordinates of the center points of all camera devices in the frame image are output to obtain the set of coordinates of the center points of the camera devices in the frame image.

[0008] If the set of camera device center point coordinates of the frame image is a non-empty set, then the frame image is input into the human skeleton extraction model to extract the coordinates of all human eyes center points in the frame image, and obtain the set of human eyes center point coordinates of the frame image.

[0009] Calculate the Euclidean distance between the coordinates of each camera center point in the set of camera center point coordinates of the image frame and the coordinates of each human eye center point in the set of human eye center point coordinates of the image frame. Use the minimum calculated Euclidean distance as the detection result of the image frame.

[0010] The detection results of each frame in the video frame to be evaluated are obtained. If the detection results of a preset number of frames in the video frame to be evaluated are all less than a preset threshold, and the preset number of frames are continuous in time, then it is determined that there is a photo-taking behavior in the video frame to be evaluated.

[0011] Preferably, after obtaining the set of coordinates of the center points of the human eyes in the frame image, the method further includes:

[0012] Based on the resolution of the target detection model, the coordinates of the center points of each human eye in the set of human eye center point coordinates are transformed so that the resolution of the transformed human eye center point coordinates is the same as the resolution of the center point coordinates of the camera device.

[0013] Preferably, the formula for converting the coordinates of the center points of the human eyes is:

[0014]

[0015]

[0016] Where, x e ′ x represents the converted x-coordinate of the center point of the human eyes. e y = x, where W1 is the x-coordinate of the center point of the human eyes before conversion, W2 is the resolution width of the human skeleton extraction model, and W3 is the resolution width of the object detection model. e ′ Let y be the transformed ordinate of the center point of the human eyes. e H1 represents the ordinate of the center point of the human eyes before conversion, H2 represents the resolution height of the human skeleton extraction model, and H3 represents the resolution height of the target detection model.

[0017] Preferably, the formula for calculating the Euclidean distance between the center point coordinates of the camera device and the center points of the human eyes is as follows:

[0018]

[0019] Among them, D ij Let x represent the Euclidean distance between the coordinates of the i-th camera center point in the set of camera center point coordinates and the coordinates of the j-th human eye center point in the set of human eye center point coordinates. pi The x-coordinate of the i-th camera center point in the set of camera center point coordinates is represented by y.pi This represents the ordinate of the i-th camera center point in the set of camera center point coordinates, where i∈[1,n] and n is the total number of camera center point coordinates in the set. ej Let y represent the x-coordinate of the j-th eye center point in the set of eye center point coordinates. ej Let represent the ordinate of the j-th eye center point in the set of eye center point coordinates, where j∈[1,m] and m is the total number of eye center point coordinates in the set of eye center point coordinates.

[0020] Preferably, after using the calculated minimum Euclidean distance as the detection result of the frame image, the method further includes:

[0021] Obtain the depth of field of the frame image, and use the ratio of the depth of field of the frame image to the preset reference depth as the adjustment factor of the frame image;

[0022] The calculated minimum Euclidean distance is updated based on the adjustment factor of the frame image, and the updated minimum Euclidean distance is used as the detection result of the frame image.

[0023] Preferably, the formula for calculating the adjustment factor is:

[0024]

[0025] Where α represents the adjustment factor, D scenn D represents depth of field. base Indicates the preset reference depth;

[0026] The formula for updating the calculated minimum Euclidean distance is:

[0027] D adjusted =α×D min ,

[0028] Among them, D adjusted D represents the updated minimum Euclidean distance. min This represents the minimum Euclidean distance calculated before the update.

[0029] Preferably, obtaining the video frame to be evaluated includes: obtaining the video to be evaluated, and performing frame segmentation processing on the video to be evaluated to obtain the video frame to be evaluated.

[0030] Preferably, if the set of coordinates of the camera center point of the frame image is empty, then other frame images in the video frame to be evaluated are input into the target detection model.

[0031] The present invention also provides a human posture assessment device, comprising:

[0032] The camera device center point coordinate acquisition module is used to acquire the video frame to be evaluated, input each frame image in the video frame to be evaluated into the target detection model, output the coordinates of all camera device center points in the frame image, and obtain the set of camera device center point coordinates of the frame image;

[0033] The human eye center point coordinate acquisition module is used to input the frame image into the human skeleton extraction model if the set of camera device center point coordinates of the frame image is a non-empty set, and extract the coordinates of all human eye center points in the frame image to obtain the set of human eye center point coordinates of the frame image.

[0034] The image detection result acquisition module is used to calculate the Euclidean distance between the coordinates of each camera center point in the set of camera center point coordinates of the image frame and the coordinates of each human eye center point in the set of human eye center point coordinates of the image frame, and to use the minimum calculated Euclidean distance as the detection result of the image frame.

[0035] The posture evaluation module is used to obtain the detection results of each frame in the video frame to be evaluated. If the detection results of a preset number of frames in the video frame to be evaluated are all less than a preset threshold, and the preset number of frames are continuous in time, then it is determined that there is a photo-taking behavior in the video frame to be evaluated.

[0036] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described human posture assessment method.

[0037] The human pose assessment method provided in this application first uses a target detection model to obtain the set of camera device center point coordinates for each frame of the video frame to be evaluated. When the set of camera device center point coordinates for that frame is a non-empty set, a human skeleton extraction model is activated to obtain the set of human eye center point coordinates for that frame. This interactive, serial working mode of the two models reduces computational load and improves assessment efficiency. Furthermore, since the probability of a human body appearing in a video frame is greater than the probability of a camera device appearing, this application first uses a target detection model to detect the camera device in the image; only when the set of camera device center point coordinates is a non-empty set is the human pose assessment method activated. Activating the human skeleton extraction model can reduce its trigger rate and resource consumption. Meanwhile, considering the small distance between the human eye and the camera device during photography, and that photography is a stable and continuous process, this application uses the minimum distance between the human eye and the camera device in each frame as the detection result for that frame. By obtaining the detection results of each frame in the video frame to be evaluated, and determining whether there exists a preset number of frames in the video frame to be evaluated whose detection results are all less than a preset threshold, and which are temporally continuous, it can accurately identify whether there is a photographing behavior in the video frame to be evaluated. Attached Figure Description

[0038] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein:

[0039] Figure 1 The flowchart of the human posture assessment method provided in this application;

[0040] Figure 2 A schematic diagram of key points of the human skeleton extracted from the human skeleton extraction model provided in this application;

[0041] Figure 3 A schematic diagram illustrating the performance metrics of the object detection model provided in this application on the training set; wherein, Figure 3 In the diagram, (a) represents the loss of the object detection model when detecting bounding boxes of target objects on the training set. Figure 3 (b) in the figure represents the accuracy of the object detection model in detecting the presence of target objects on the training set. Figure 3 In the diagram, (c) represents the classification accuracy of the object detection model on the training set. Figure 3 In the figure, (d) represents the proportion of object categories correctly predicted by the object detection model on the training set. Figure 3 In this context, (e) represents the recall rate of the object detection model on the training set.

[0042] Figure 4 A schematic diagram illustrating the performance metrics of the target detection model provided in this application on the validation set; wherein, Figure 4 In the diagram, (a) represents the loss of the object detection model when detecting the bounding box of an object on the validation set. Figure 4 (b) in the figure represents the accuracy of the object detection model in detecting the presence of the target object on the validation set. Figure 4 In the diagram, (c) represents the classification accuracy of the object detection model on the validation set. Figure 4 In the diagram, (d) represents the map value calculated by the object detection model when the intersection-union ratio threshold is 0.5. Figure 4 In the figure, (e) represents the average map value calculated by the target detection model at different crossover ratio thresholds (0.5 to 0.95);

[0043] Figure 5 A schematic diagram of the human posture assessment device provided in this application. Detailed Implementation

[0044] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0045] Please see Figure 1, Figure 1 The figure shown is a human posture assessment method provided in this application, which specifically includes:

[0046] S10: Obtain the video frame to be evaluated, input each frame image in the video frame to be evaluated into the target detection model, output the coordinates of the center points of all camera devices in the frame image, and obtain the set of coordinates of the center points of the camera devices in the frame image;

[0047] Alternatively, the camera device can be a mobile phone, camcorder, or other device that can be used to take pictures.

[0048] S20: If the set of camera device center point coordinates of the frame image is a non-empty set, then input the frame image into the human skeleton extraction model, extract the coordinates of all human eyes center points in the frame image, and obtain the set of human eyes center point coordinates of the frame image.

[0049] S30: Calculate the Euclidean distance between the coordinates of each camera center point in the set of camera center point coordinates of the frame image and the coordinates of each human eye center point in the set of human eye center point coordinates of the frame image, and take the minimum Euclidean distance as the detection result of the frame image.

[0050] S40: Obtain the detection result of each frame in the video frame to be evaluated. If the detection results of a preset number of frames in the video frame to be evaluated are all less than a preset threshold, and the preset number of frames are continuous in time, then it is determined that there is a photo-taking behavior in the video frame to be evaluated.

[0051] Specifically, the human skeletal body region model focuses on identifying and tracking the relative positions of various parts of the human body, such as... Figure 2The diagram shows a human skeleton extracted by the human skeleton extraction model provided in this application. The object detection model focuses on identifying specific objects in the image. Since these two processes are performed independently, they require additional processing time and computational resources, thus reducing overall efficiency and accuracy. Therefore, this application first uses the object detection model to obtain the set of camera device center point coordinates for each frame of the video frame to be evaluated. When the set of camera device center point coordinates for that frame is non-empty, the human skeleton extraction model is activated to obtain the set of human eye center point coordinates for that frame. This interactive, serial working mode of the two models reduces computation and improves evaluation efficiency. Furthermore, since the probability of a human body appearing in a video frame is greater than the probability of a camera device appearing, this application… First, the object detection model is used to detect the camera device in the image. The human skeleton extraction model is only activated when the coordinate set of the camera device's center point is a non-empty set. This reduces the trigger rate of the human skeleton extraction model and decreases resource consumption. At the same time, considering that the distance between the human eye and the camera device is small during photography and that photography is a stable and continuous process, this application uses the minimum distance between the human eye and the camera device in each frame as the detection result of that frame. By obtaining the detection results of each frame in the video frame to be evaluated, and determining whether there is a preset number of frames in the video frame to be evaluated whose detection results are all less than a preset threshold, and the preset number of frames are continuous in time, it is possible to accurately identify whether there is a photography behavior in the video frame to be evaluated.

[0052] Specifically, step S10, obtaining the video frame to be evaluated, includes: obtaining the video to be evaluated and performing frame segmentation processing on the video to be evaluated to obtain the video frame to be evaluated.

[0053] In step S20, if the set of camera device center point coordinates of the frame image is empty, then other frame images in the video frame to be evaluated are input into the target detection model.

[0054] Preferably, since the target detection model and the human skeleton extraction model have different resolutions, after obtaining the set of coordinates of the center points of the human eyes, it is also necessary to map the coordinates of the center points of the human eyes to the same resolution environment as the camera device center point coordinates output by the target detection model.

[0055] Specifically, step S20 is followed by:

[0056] Based on the resolution of the target detection model, the coordinates of the center points of each human eye in the set of human eye center point coordinates are transformed so that the resolution of the transformed human eye center point coordinates is the same as the resolution of the center point coordinates of the camera device.

[0057] Specifically, the formula for converting the coordinates of the center points of the human eyes is:

[0058]

[0059]

[0060] Where, x e ′ x represents the converted x-coordinate of the center point of the human eyes. e y = x, where W1 is the x-coordinate of the center point of the human eyes before conversion, W2 is the resolution width of the human skeleton extraction model, and W3 is the resolution width of the object detection model. e ′ Let y be the transformed ordinate of the center point of the human eyes. e H1 represents the ordinate of the center point of the human eyes before conversion, H2 represents the resolution height of the human skeleton extraction model, and H3 represents the resolution height of the target detection model.

[0061] For example, in a specific example of this application, the output resolution of the target detection model is 1280*720, while the output resolution of the serially running human skeleton extraction model is 960*576. Therefore, this application can convert the coordinates of the camera center point and the coordinates of the human eyes center point to the same resolution through coordinate transformation, thereby making the calculated Euclidean distance between the two more accurate.

[0062] Specifically, there may be multiple camera devices and human bodies in the same frame of an image. Considering that the distance between the human eyes and the camera devices is small during the photo taking process, this application only uses the minimum distance between the human eyes and the camera devices calculated in one frame of an image as the detection result of that frame of an image.

[0063] Specifically, the formula for calculating the Euclidean distance between the center point coordinates of the camera device and the center points of the human eyes in step S30 is as follows:

[0064]

[0065] Among them, D ij Let x represent the Euclidean distance between the coordinates of the i-th camera center point in the set of camera center point coordinates and the coordinates of the j-th human eye center point in the set of human eye center point coordinates. pi The x-coordinate of the i-th camera center point in the set of camera center point coordinates is represented by y. pi This represents the ordinate of the i-th camera center point in the set of camera center point coordinates, where i∈[1,n] and n is the total number of camera center point coordinates in the set. ej Let y represent the x-coordinate of the j-th eye center point in the set of eye center point coordinates. ej Let represent the ordinate of the j-th eye center point in the set of eye center point coordinates, where j∈[1,m] and m is the total number of eye center point coordinates in the set of eye center point coordinates.

[0066] After calculating the detection result of a frame of image, this embodiment of the application also eliminates depth interference by proportional compression, thereby obtaining a more accurate distance between the center point of human eyes and the center point of the camera device, which greatly enriches the robustness of the data.

[0067] Specifically, step S30 is followed by:

[0068] Obtain the depth of field of the frame image, and use the ratio of the depth of field of the frame image to the preset reference depth as the adjustment factor of the frame image;

[0069] The calculated minimum Euclidean distance is updated based on the adjustment factor of the frame image, and the updated minimum Euclidean distance is used as the detection result of the frame image.

[0070] Specifically, the formula for calculating the adjustment factor is as follows:

[0071]

[0072] Where α represents the adjustment factor, D scenn D represents depth of field. base Indicates the preset reference depth;

[0073] The formula for updating the calculated minimum Euclidean distance is:

[0074] D adjusted =α×D min ,

[0075] Among them, D adjusted D represents the updated minimum Euclidean distance. min This represents the minimum Euclidean distance calculated before the update.

[0076] Depth of field refers to the depth information extracted from an image. For example, if an image contains a standard-sized object, the depth of field can be estimated by the ratio of the object's size in the image to its actual size in reality. Reference depth refers to a predefined reference depth value used in a specific application or environment to calibrate or compare other depth measurements. Reference depth helps in understanding and correcting potential biases or errors in depth perception.

[0077] The methods for obtaining the baseline depth can vary depending on the application scenario, for example:

[0078] 1. Pre-measurement: In a known environment, the depth of a specific point or object can be pre-measured using standard measuring tools (such as laser rangefinders and rulers), and the pre-measured depth value can be used as the reference depth for the scene.

[0079] 2. Use standard objects: If there are standard objects of known size in the scene (such as standard-sized doors or furniture), the baseline depth can be estimated by comparing the size of these objects in the image with their actual size.

[0080] 3. Software simulation or preset: A baseline depth value is preset or calculated in virtual reality or computer simulation.

[0081] 4. Historical data analysis: In some cases, the baseline depth can be determined by analyzing historical data.

[0082] When adjusting the detection results of an image, this application can more accurately understand and adjust the measurement errors caused by factors such as depth of field and perspective distortion by comparing the depth of field with the reference depth.

[0083] Specifically, in some embodiments of this application, in order to improve the ability of the object detection model and the human skeleton extraction model to recognize objects of different sizes, a multi-scale training strategy can be used to train the model. That is, before training, the images in the training set are adjusted to different sizes. For example, an image scaling algorithm can be used to adjust the images to sizes such as 256*256, 512*512, and 1024*1024. The model is trained with images of different sizes in different training batches, so that the model can extract effective information from images of different sizes and improve the generalization ability of the model.

[0084] Please see Figure 3 and Figure 4 These are schematic diagrams illustrating the performance metrics of the object detection model provided in this application on the training and validation sets, respectively. Figure 3 and Figure 4 In the diagram, (a) represents the loss of the object detection model when detecting bounding boxes of target objects on the training and validation sets, respectively. Figure 3 and Figure 4 In the diagram, (b) represents the accuracy of the object detection model in detecting the presence of target objects on the training and validation sets, respectively. Figure 3 and Figure 4 In the diagram, (c) represents the classification accuracy of the object detection model on the training and validation sets, respectively. Figure 3 In the diagram, (d) represents the proportion of object categories correctly predicted by the object detection model on the training set. Figure 3 In this context, (e) represents the recall rate of the object detection model on the training set. Figure 4 In the diagram, (d) represents the map value calculated by the object detection model when the intersection-union ratio (IU) threshold is 0.5, indicating the model's detection accuracy and localization accuracy. Figure 4 In the figure, (e) represents the average map value calculated by the object detection model at different crossover ratio thresholds (0.5 to 0.95).

[0085] Based on the human posture assessment method provided in the above embodiments, this application also provides a human posture assessment device, such as... Figure 5 As shown, the device specifically includes:

[0086] The camera device center point coordinate acquisition module 10 is used to acquire the video frame to be evaluated, input each frame image in the video frame to be evaluated into the target detection model, output the coordinates of all camera device center points in the frame image, and obtain the set of camera device center point coordinates of the frame image.

[0087] The human eye center point coordinate acquisition module 20 is used to input the frame image into the human skeleton extraction model if the set of camera device center point coordinates of the frame image is a non-empty set, and extract all human eye center point coordinates in the frame image to obtain the human eye center point coordinate set of the frame image.

[0088] The image detection result acquisition module 30 is used to calculate the Euclidean distance between the coordinates of each camera center point in the set of camera center point coordinates of the frame image and the coordinates of each human eye center point in the set of human eye center point coordinates of the frame image, and to use the minimum calculated Euclidean distance as the detection result of the frame image.

[0089] The posture evaluation module 40 is used to obtain the detection results of each frame in the video frame to be evaluated. If the detection results of a preset number of frames in the video frame to be evaluated are all less than a preset threshold, and the preset number of frames are continuous in time, then it is determined that there is a photo-taking behavior in the video frame to be evaluated.

[0090] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described human posture assessment method.

[0091] The human pose assessment method provided in this application significantly improves processing speed by integrating an efficient target detection model and a human skeleton extraction model, enabling real-time human pose assessment. This is of great significance for application scenarios requiring rapid response, such as real-time monitoring and interactive systems. Furthermore, it maintains high accuracy even in complex backgrounds, varying lighting conditions, or partial occlusion, significantly improving the applicability and reliability of this method in various environments. In addition, by training the target detection model and the human skeleton extraction model through a multi-scale training strategy, the model can effectively handle the detection of camera devices of different sizes and various visual tasks simultaneously, adapting to a wider range of application needs.

[0092] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0093] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0094] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for assessing human posture, characterized in that, include: Obtain the video frame to be evaluated, input each frame image in the video frame to be evaluated into the target detection model, output the coordinates of the center points of all camera devices in the frame image, and obtain the set of coordinates of the center points of the camera devices in the frame image; If the set of camera device center point coordinates of the frame image is a non-empty set, then the frame image is input into the human skeleton extraction model to extract the coordinates of all human eyes center points in the frame image, and the set of human eyes center point coordinates of the frame image is obtained. Based on the resolution of the target detection model, the coordinates of the center points of each human eye in the set of human eye center point coordinates are transformed so that the resolution of the transformed human eye center point coordinates is the same as the resolution of the center point coordinates of the camera device. Calculate the Euclidean distance between the coordinates of each camera center point in the set of camera center point coordinates of the image frame and the coordinates of each human eye center point in the set of human eye center point coordinates of the image frame. Use the minimum calculated Euclidean distance as the detection result of the image frame. Obtain the depth of field of the frame image, and use the ratio of the depth of field of the frame image to the preset reference depth as the adjustment factor of the frame image; The minimum Euclidean distance is updated based on the adjustment factor of the frame image, and the updated minimum Euclidean distance is used as the detection result of the frame image; wherein, the formula for calculating the adjustment factor is: ， in, Indicates the regulating factor. Indicates depth of field. Indicates the preset reference depth; The formula for updating the calculated minimum Euclidean distance is: ， in, This represents the updated minimum Euclidean distance. This represents the minimum Euclidean distance calculated before the update; The detection results of each frame in the video frame to be evaluated are obtained. If the detection results of a preset number of frames in the video frame to be evaluated are all less than a preset threshold, and the preset number of frames are continuous in time, then it is determined that there is a photo-taking behavior in the video frame to be evaluated.

2. The human posture assessment method according to claim 1, characterized in that, The formula for converting the coordinates of the center points of the human eyes is: ， ， in, The x-coordinates of the center points of the human eyes after conversion. The x-coordinates of the center points of the human eyes before conversion. Extract the resolution width of the human skeleton model. This represents the resolution width of the target detection model. The transformed ordinates are the vertical coordinates of the center points of the human eyes. The coordinates of the center points of the human eyes before conversion are: Extract the resolution height of the human skeleton model. This represents the resolution height of the target detection model.

3. The human posture assessment method according to claim 1, characterized in that, The formula for calculating the Euclidean distance between the center point of the camera device and the center points of the human eyes is: ， in, Represents the set of coordinates of the center point of the camera device. In the set of coordinates of the center point of the camera device and the center point coordinates of the human eyes, the first... The Euclidean distance between the coordinates of the center points of an individual's two eyes. Represents the set of coordinates of the center point of the camera device. The x-coordinate of the center point of each camera device Represents the set of coordinates of the center point of the camera device. The ordinate of the center point of each camera device , The total number of camera center point coordinates in the set of camera center point coordinates. Represents the set of coordinates of the center points of the human eyes. The x-coordinate of the center point of each person's eyes. Represents the set of coordinates of the center points of the human eyes. The ordinate of the center point of each person's eyes. , This represents the total number of coordinates of the center points of both eyes in the set of coordinates of the human eyes.

4. The human posture assessment method according to claim 1, characterized in that, Obtaining the video frame to be evaluated includes: obtaining the video to be evaluated, and performing frame segmentation processing on the video to be evaluated to obtain the video frame to be evaluated.

5. The human posture assessment method according to claim 1, characterized in that, If the set of coordinates of the camera center point of the frame image is empty, then other frame images in the video frame to be evaluated are input into the target detection model.

6. A human posture assessment device, characterized in that, include: The camera device center point coordinate acquisition module is used to acquire the video frame to be evaluated, input each frame image in the video frame to be evaluated into the target detection model, output the coordinates of all camera device center points in the frame image, and obtain the set of camera device center point coordinates of the frame image; The human eye center point coordinate acquisition module is used to input the frame image into the human skeleton extraction model if the set of camera device center point coordinates of the frame image is a non-empty set, and extract the coordinates of all human eye center points in the frame image to obtain the set of human eye center point coordinates of the frame image. Based on the resolution of the target detection model, the coordinates of the center points of each human eye in the set of human eye center point coordinates are transformed so that the resolution of the transformed human eye center point coordinates is the same as the resolution of the center point coordinates of the camera device. The image detection result acquisition module is used to calculate the Euclidean distance between the coordinates of each camera center point in the set of camera center point coordinates of the image frame and the coordinates of each human eye center point in the set of human eye center point coordinates of the image frame, and to use the minimum calculated Euclidean distance as the detection result of the image frame. Obtain the depth of field of the frame image, and use the ratio of the depth of field of the frame image to the preset reference depth as the adjustment factor of the frame image; The minimum Euclidean distance is updated based on the adjustment factor of the frame image, and the updated minimum Euclidean distance is used as the detection result of the frame image; wherein, the formula for calculating the adjustment factor is: ， in, Indicates the regulating factor. Indicates depth of field. Indicates the preset reference depth; The formula for updating the calculated minimum Euclidean distance is: ， in, This represents the updated minimum Euclidean distance. This represents the minimum Euclidean distance calculated before the update; The posture evaluation module is used to obtain the detection results of each frame in the video frame to be evaluated. If the detection results of a preset number of frames in the video frame to be evaluated are all less than a preset threshold, and the preset number of frames are continuous in time, then it is determined that there is a photo-taking behavior in the video frame to be evaluated.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the human posture assessment method according to any one of claims 1-5.

Images