Behavior discrimination method and device, computer equipment and storage medium

Abstract

The embodiment of the invention belongs to the field of artificial intelligence and action recognition, and relates to a behavior discrimination method, which comprises the following steps: acquiring a target monitoring image sequence, carrying out holding detection processing on target monitoring images, generating holding detection images corresponding to the target monitoring images, carrying out content discrimination processing on the holding detection images, and judging whether the holding detection images correspond to the target monitoring images or not. The method comprises the following steps: acquiring a holding detection picture to obtain a content judgment result, performing joint detection processing on the holding detection picture to obtain an action judgment result, and determining a target shooting behavior based on the content judgment result and the action judgment result. The invention further provides a behavior judgment device, computer equipment and a storage medium. According to the invention, the method can achieve the automatic and refined human stealing behavior in a fixed scene through the content discrimination result and the action discrimination result, thereby accurately and rapidly providing an early warning capability for the leakage of privacy information.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence and action recognition technology, and in particular to behavior identification methods, devices, computer equipment and storage media. Background Art

[0002] In financial trading venues such as banks and securities institutions, large amounts of sensitive data, including account information, transaction amounts, and investment details, are often processed. Any leakage of sensitive data may lead to significant financial losses and reputational crises. For example, some users may deliberately take photos of documents or computer or tablet screens at bank counters. Therefore, ensuring the security of customer transaction information and privacy is particularly critical.

[0003] Traditional prevention methods typically use physical measures such as surveillance cameras, alarm systems, and strict management of personnel entry and exit to prevent unauthorized access. However, in environments with dense traffic, physical prevention measures have limited monitoring capabilities and cannot effectively and accurately identify users' voyeuristic photography behavior in real time based on their behavior. Summary of the Invention

[0004] The purpose of the embodiments of this application is to propose a behavior identification method, device, computer equipment and storage medium, the main purpose of which is to accurately identify potential illegal shooting actions by real-time monitoring and analysis of individual behaviors within the monitoring range, thereby improving the ability to prevent the leakage of sensitive information.

[0005] In order to solve the above technical problems, the present application provides a behavior identification method, which adopts the following technical solutions:

[0006] Acquire a target monitoring image sequence, wherein the target monitoring image sequence includes at least one target monitoring picture;

[0007] Performing grip detection processing on the target monitoring pictures to generate a grip detection picture corresponding to each target monitoring picture;

[0008] performing content discrimination processing on the grip detection image to obtain a content discrimination result;

[0009] Performing joint detection processing on the grip detection image to obtain an action discrimination result;

[0010] A target shooting behavior is determined based on the content identification result and the action identification result.

[0011] Furthermore, before the step of performing grip detection processing on the target monitoring picture to generate a grip detection picture corresponding to each target monitoring picture, the method further includes:

[0012] Pulling a grip picture training dataset and a grip picture test dataset from a historical image database, wherein the grip picture training dataset includes at least one grip training picture, and the grip picture test dataset includes at least one grip test picture;

[0013] Performing data preprocessing on the grip training image to obtain target grip feature parameters;

[0014] Inputting the target grip feature parameters into a preset universal model, and training the preset universal model using the grip image training dataset to obtain a grip detection model;

[0015] Importing the grip image test dataset into the grip detection model and outputting the detection results;

[0016] Based on the detection result and the preset standard result, a back propagation algorithm is used to perform fitting to obtain a loss error, and the grip detection model is updated based on the loss error.

[0017] Furthermore, the grip detection model includes an input layer, a prediction layer, and an output layer. The step of performing grip detection processing on the target monitoring image to generate a grip detection image corresponding to each target monitoring image specifically includes:

[0018] Inputting the target monitoring image into the input layer to perform boundary prediction and obtain boundary confidence;

[0019] Inputting the boundary confidence into the prediction layer for category screening to obtain a boundary prediction result;

[0020] The boundary prediction result is input into the output layer for image cropping, and a grip detection image corresponding to the target monitoring image is output.

[0021] Furthermore, before the step of performing content identification processing on the grip detection image to obtain a content identification result, the following steps are specifically included:

[0022] Pulling a screen content training data set and a screen content test data set from the historical image database, wherein the screen content training data set includes at least one screen content training image, and the screen content test data set includes at least one screen content test image;

[0023] Performing data preprocessing on the screen content training graph to obtain target content feature parameters;

[0024] Inputting the target content feature parameters into a preset universal model, and training the preset universal model using the screen content training data set to obtain a content detection model;

[0025] Importing the screen content test data set into the content detection model and outputting the content detection result;

[0026] Based on the content detection result and the preset standard result, a back propagation algorithm is used to perform fitting to obtain a content loss error, and the content detection model is updated based on the content loss error.

[0027] Furthermore, the content detection model includes a bright screen detection layer and a content recognition layer. The step of performing content discrimination processing on the holding detection image to obtain a content discrimination result specifically includes:

[0028] Inputting the holding detection picture into the screen-on detection layer to perform screen-on detection and obtain a screen-on detection result;

[0029] The bright screen detection result is used as the input of the content recognition layer to perform content identification processing to obtain a content discrimination result.

[0030] Furthermore, the step of performing joint detection processing on the grip detection image to obtain an action discrimination result further includes:

[0031] Performing hand detection processing on the grip detection image to obtain a target hand image;

[0032] Performing joint deconstruction on the target hand graph to obtain a target joint structure graph;

[0033] The target joint structure diagram is subjected to bending angle detection to obtain the action discrimination result.

[0034] Furthermore, the target joint structure diagram includes a thumb joint and an index finger joint, and the step of performing bending angle detection on the target joint structure diagram to obtain the action discrimination result specifically includes:

[0035] Calculating the bending angle difference of the thumb joint to obtain the curvature of the thumb;

[0036] Calculating the bending angle difference of the index finger joint to obtain the index finger curvature;

[0037] The action discrimination result is determined based on the thumb curvature and the thumb bending degree.

[0038] In order to solve the above technical problems, the present application also provides a behavior identification device, which adopts the following technical solutions:

[0039] An acquisition module is used to acquire a target monitoring image sequence, wherein the target monitoring image sequence includes at least one target monitoring picture;

[0040] a grip detection module, configured to perform grip detection processing on the target monitoring image and generate a grip detection image corresponding to each target monitoring image;

[0041] a discrimination module, configured to perform content discrimination processing on the grip detection image to obtain a content discrimination result;

[0042] A joint detection module is used to perform joint detection processing on the grip detection image to obtain an action discrimination result;

[0043] The determination module is configured to determine the target shooting behavior based on the content identification result and the action identification result.

[0044] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0045] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the behavior identification method when executing the computer-readable instructions.

[0046] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0047] A computer-readable storage medium stores computer-readable instructions, which implement the steps of the behavior identification method when executed by a processor.

[0048] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0049] The embodiments of the present application obtain a target monitoring image sequence, perform grip detection processing on the target monitoring images, generate a grip detection image corresponding to each target monitoring image, perform content discrimination processing on the grip detection images to obtain a content discrimination result, perform joint detection processing on the grip detection images to obtain an action discrimination result, determine the target shooting behavior based on the content discrimination result and the action discrimination result, and accurately identify potential illegal shooting actions by real-time monitoring and analysis of the individual behavior of users within the monitoring range, as well as combining the specific display content of the user device, thereby improving the prevention capability against leakage of sensitive information. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the solutions in this application, a brief introduction will be given below to the drawings required for use in the description of the embodiments of this application. Obviously, the drawings described below are some embodiments of this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0051] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0052] Figure 2 A flowchart of an embodiment of a behavior identification method according to the present application;

[0053] Figure 3 is a structural diagram of an embodiment of a behavior identification device according to the present application;

[0054] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of the application are for the purpose of describing specific embodiments only and are not intended to limit this application. The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0056] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0057] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0058] like Figure 1As shown, system architecture 100 may include a terminal device 101, a network 102, and a server 103. Terminal device 101 may be a laptop computer 1011, a tablet computer 1012, or a mobile phone 1013. Network 102 is a medium for providing a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0059] The user can use the terminal device 101 to interact with the server 103 via the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0060] The terminal device 101 can be various electronic devices with a display screen and supporting web browsing. In addition to the laptop computer 1011, tablet computer 1012 or mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV) player, a laptop computer and a desktop computer, etc.

[0061] The server 103 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal device 101 .

[0062] It should be noted that the behavior identification method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the behavior identification device is generally set in the server / terminal device.

[0063] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0064] Continue to refer Figure 2 , which shows a flow chart of an embodiment of a method for behavior identification according to the present application. The behavior identification method comprises the following steps:

[0065] Step S201: Acquire a target monitoring image sequence.

[0066] In this embodiment, the above-mentioned behavior discrimination method can be deployed in a behavior discrimination platform. The above-mentioned behavior discrimination system can be constructed by a server or a server cluster. The above-mentioned server or server cluster can be any electronic device with image recognition, image processing, text processing, text recognition, data transmission, and data storage functions. The behavior discrimination method runs on the electronic device thereon (for example, Figure 1 The server / terminal device shown in the figure can receive the target monitoring image sequence via a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0067] In this embodiment, the target surveillance image sequence includes at least one target surveillance picture. Specifically, the target surveillance image sequence may be a series of surveillance images captured continuously or discontinuously, and the target surveillance picture may be a single image frame in the target surveillance image sequence. The target surveillance picture may also be a single photo captured by a high-definition camera or a static camera.

[0068] In one possible embodiment, the target surveillance image sequence may be a series of images captured by a surveillance camera in a public trading venue, such as a bank or stock exchange. The images are recorded every few seconds to capture the activities of customers and staff within the public trading venue. For example, in a 10-minute video, the surveillance camera captures a frame every 2 seconds for analyzing behavioral patterns, such as movement, interaction, or lingering in a specific area. The target surveillance image may be a surveillance image captured by a high-definition camera, which may show the customer holding a mobile phone, possibly taking a photo.

[0069] Step S202 : performing grip detection processing on the target monitoring pictures to generate a grip detection picture corresponding to each target monitoring picture.

[0070] In this embodiment, the above-mentioned holding detection picture can be a picture used to highlight or identify the user's hand-holding action or holding state in the image, and the above-mentioned holding detection picture can be used to represent and mark the relationship between the user's hand and the object held (such as a mobile phone).

[0071] In one possible embodiment, a surveillance camera in a bank lobby captures a customer standing in front of a counter, clearly holding a smartphone and operating the phone. At this time, the above-mentioned behavior recognition system will control the surveillance camera to capture an image of the customer at the current time point and clearly mark the position of the hand holding the phone, such as using a border or highlighting the area of ​​the hand and the phone, as the above-mentioned holding detection image.

[0072] Specifically, regarding the specific implementation process of performing grip detection processing on the target monitoring image and generating a grip detection image corresponding to each target monitoring image, this application will provide further details in subsequent specific embodiments and will not be elaborated on here.

[0073] Step S203: performing content identification processing on the grip detection image to obtain a content identification result.

[0074] In this embodiment, the content identification result may be the text, image, or other visual elements displayed on the mobile phone screen in the grip detection image, for example, the screen display of a mobile phone application such as a camera interface, a map, a game, or business software on the user's mobile phone.

[0075] Specifically, regarding the specific implementation process of performing content discrimination processing on the grip detection image and obtaining the content discrimination result, this application will provide further details in subsequent specific embodiments and will not be elaborated on here.

[0076] Step S204: perform joint detection processing on the grip detection image to obtain an action discrimination result.

[0077] In this embodiment, the above-mentioned action identification result can be the behavioral intention and action type determined by analyzing the hand joint position and movement in the grip detection image. Specifically, the above-mentioned action identification result can include legal behavior or sensitive behavior. The above-mentioned legal behavior can include but is not limited to using a finger to slide the screen up and down or left and right to browse the web or read electronic documents, tapping the screen to select an application or option, using a virtual keyboard to input text, etc.; the above-mentioned sensitive behavior can include but is not limited to using a mobile phone camera to point at a computer screen or document, pointing a mobile phone camera at a specific object or document for a long time, and starting and operating software applications that may be used to steal information or invade the system.

[0078] Specifically, regarding the specific implementation process of performing joint detection processing on the grip detection image to obtain the action discrimination result, this application will provide further details in subsequent specific embodiments and will not be elaborated on here.

[0079] Step S205 : determining the target shooting behavior based on the content identification result and the action identification result.

[0080] In this embodiment, the target photography behavior can be a user's sensitive photography behavior or normal photography behavior. Specifically, the above-mentioned sensitive photography behavior includes but is not limited to using a camera inside a bank's confidential area or other places where photography is explicitly prohibited, using a camera to record personal bank card details, sensitive transaction information displayed on the screen, or other customer personal information, and using a mobile phone camera to record a sensitive area for a long time. The above-mentioned normal photography behavior includes but is not limited to photography of vegetation, scenery, and taking photos with others with their permission.

[0081] In a possible embodiment, the above-mentioned behavior identification system detects through a surveillance camera the user's behavior of using a camera inside a confidential area of ​​a bank or other places where taking photos is explicitly prohibited (i.e., the above-mentioned action identification result), and the user's mobile device displays bank account details, credit card number, and ongoing transactions or transaction confirmation pages (i.e., the above-mentioned content identification result), and then determines that the current user's target shooting behavior is a sensitive behavior.

[0082] The present application obtains a target monitoring image sequence, performs grip detection processing on the target monitoring pictures, generates a grip detection picture corresponding to each target monitoring picture, performs content discrimination processing on the grip detection pictures to obtain a content discrimination result, performs joint detection processing on the grip detection pictures to obtain an action discrimination result, determines the target shooting behavior based on the content discrimination result and the action discrimination result, and accurately identifies potential illegal shooting actions by real-time monitoring and analysis of the individual behavior of users within the monitoring range, and combining the specific display content of the user's device, thereby improving the prevention capability of sensitive information leakage.

[0083] In some optional implementations, before step S202, the following steps are further included:

[0084] Pull out the grip image training dataset and grip image test dataset from the historical image database;

[0085] Perform data preprocessing on the grip training image to obtain target grip feature parameters;

[0086] Input the target grip feature parameters into a preset universal model, and train the preset universal model using a grip image training dataset to obtain a grip detection model;

[0087] Import the grip image test dataset into the grip detection model and output the detection results;

[0088] Based on the detection results and the preset standard results, the back propagation algorithm is used for fitting to obtain the loss error, and the grip detection model is updated based on the loss error.

[0089] In this embodiment, the historical image database may be a large image storage repository formed by aggregating multiple data sources. Specifically, the large image storage repository may be a bank's internal surveillance image database.

[0090] In this embodiment, the grip image training dataset includes at least one grip training image, and the grip image test dataset includes at least one grip test image. Specifically, the grip image training dataset may include, but is not limited to, a large number of images of people holding different objects (such as mobile phones, books, tools, etc.) in different environments collected from surveillance cameras, smartphones, or other image capture devices. The grip image test dataset may be an independent image set separate from the grip image training dataset, used to evaluate the effectiveness of the grip detection model.

[0091] In this embodiment, the target grip feature parameters may include, but are not limited to, the position of hand joints, finger spread, the angle of gripping the object, and a pressure map of the contact points between the hand and the object. Specifically, the target grip feature parameters may be extracted from a grip image training dataset and used to train the grip detection model to identify features of the hand and the gripped object.

[0092] In a possible embodiment, regarding the acquisition of a gripping picture training dataset and a gripping picture test dataset, a total of 580 images can be downloaded from the Internet, and 4,308 images are collected from the company's internal bank business hall scene, wherein each image may contain one or more obvious mobile phone screens. All these images are saved in the same folder, and then the open source tool labelImg is used to annotate these images. The annotation process involves framing the mobile phone screen on each image and giving it an object name, such as "corner". The result of each annotation is saved in XML format, which contains a custom image object name (such as "photo1"), image file name, file path, image size (width, height, depth), and the coordinates of the handheld mobile phone screen area (minimum x coordinate, minimum y coordinate, maximum x coordinate, maximum y coordinate), and The object name of the annotated mobile phone screen, all XML files are organized in another special folder to ensure that each image file and its corresponding XML file correspond one-to-one through the file name, which is convenient for subsequent processing (i.e., the above-mentioned data preprocessing). In the process of dividing the data set, the ratio of the training set to the validation set is 9:1, and the ratio of the training set to the test set is 9:1. After determining the number of divisions, we randomly extract the XML file to obtain the image file name, remove the file name suffix such as .jpg, and save it as an index to a txt file. The above txt files can include test.txt, train.txt, trainval.txt and val.txt. The index can access the annotation content and the original image in the XML at the same time, thereby providing a structured and efficient data access method for grip detection model training.

[0093] In this embodiment, the general model can be a convolutional neural network (CNN) based on deep learning, designed to recognize and analyze holding behaviors in images. The general model can adopt existing network architectures such as ResNet, VGG or Inception.

[0094] In a possible embodiment, the above-mentioned grasping detection model can adopt Darknet53 based on the YOLO framework as the backbone network, and add a CSP structure to each large residual block thereof, covering from layer0 to layer104. In order to expand the receptive field of the network, the spatial pyramid pooling technology is introduced, and layer107 is subjected to maximum pooling of 5x5, 9x9, and 13x13 to generate layer108, ​​layer110 and layer112. After the pooling is completed, the above layers are merged into the feature map layer114, and the dimension is reduced to 512 channels through 1x1 convolution. At the same time, based on the FPN architecture, effective feature fusion is achieved through upsampling and downsampling operations.

[0095] Specifically, the above Darknet53 can be a deep learning neural network, especially a convolutional neural network (CNN) architecture for object detection tasks. The above layer represents "layer", which is one of the basic units of the neural network.

[0096] In one possible embodiment, the grip detection model is trained on a system equipped with a V100 quad-GPU system. Based on the video memory capacity, the input image is resized to 416x416 pixels. To optimize the training effect, Mosaic data augmentation, label smoothing, and the CIOU loss function are used. A cosine annealing strategy is used to adjust the learning rate and the Mish activation function. Furthermore, to accelerate training and avoid weight corruption in the early stages of training, the features of the backbone feature extraction network are frozen. The entire training process lasts for 200 epochs, of which the learning rate for the first 100 epochs is set to 1e-3 and the batch size is 8. For the last 100 epochs, the learning rate is adjusted to 1e-4 and the batch size is 4 to increase training speed and reduce video memory usage, ensuring efficient and stable model training.

[0097] This application extracts a gripping picture training dataset and a gripping picture test dataset from a historical image database, preprocesses the gripping picture training dataset to extract key gripping feature parameters, and then uses the gripping feature parameters to train a gripping detection model. After the gripping detection model is trained, its performance is evaluated on the test dataset, and the model is optimized using the backpropagation algorithm to reduce prediction errors and improve detection accuracy.

[0098] In some optional implementations, step S202 includes the following steps:

[0099] Input the target monitoring image into the input layer for boundary prediction and obtain boundary confidence;

[0100] The boundary confidence is input into the processing layer for category screening to obtain the boundary prediction result;

[0101] The boundary prediction results are input into the output layer for image cropping, and the grip detection map corresponding to the target monitoring image is output.

[0102] In this embodiment, the boundary confidence may be the degree of certainty of the object boundary identified by the grip detection model. The boundary confidence is a probability value indicating how likely it is that the bounding box predicted by the grip detection model contains the target object.

[0103] In this embodiment, the boundary prediction result may include the exact location of the bounding box (usually represented by coordinates in the image) and the category of the object within the box. The boundary prediction result may be used to reflect the location and size of the object.

[0104] In a possible embodiment, the marked mobile phone screen area data can be extracted from the above-mentioned XML folder, including coordinate information such as the minimum x coordinate, minimum y coordinate, maximum x coordinate, and maximum y coordinate of each area. The above-mentioned coordinate information can be used to calculate the width and height of each bounding box. These size data are then used as input features for K-means clustering. Due to the diversity of image sizes, the sizes of the bounding boxes are also different. Therefore, the width and height of the bounding boxes need to be standardized to adapt to images of different sizes and ensure the consistency and accuracy of clustering analysis. In the K-means clustering process, the intersection over union (IOU) is used as a metric to evaluate the similarity between the cluster center (i.e., anchor) and each bounding box. When calculating the IOU, the absolute size of the bounding box is not directly considered, but is based on the normalized width and height. This processing method makes the IOU value between the bounding box and the nearest anchor larger, indicating that the shape and proportion of the bounding box are more similar, so that the bounding box can be more accurately assigned to the appropriate cluster center.

[0105] In one possible embodiment, the boundary confidence can indicate the possibility of an object being contained in the bounding box and the accuracy of its location, that is, whether the bounding box properly surrounds the mobile phone screen area. The calculation formula is:

[0106] Pr(hid)×IOU

[0107] Among them, IOU represents the intersection-over-union ratio between the predicted bounding box and the true bounding box, and Pr(hid)×IOU of each label is set to 1. In the prediction stage, the boundary confidence is used as the prediction value to represent the above meaning.

[0108] When predicting the mobile phone screen (for example, the object is named "corner"), the conditional probability value related to the specific category is also calculated, that is, the category probability under the boundary confidence. Therefore, the final score is the product of the boundary confidence and the category probability. First, a preliminary screening is performed by checking that the category probability is greater than the preset threshold of 0.5. The screened results are sorted by the size of the product of confidence and category probability. During the sorting process, the non-maximum suppression (NMS) technology is applied to remove the results with a degree of overlap greater than the preset parameter of 0.4 to obtain the optimal prediction result for each category. The above optimal prediction result is cropped according to the bounding box to obtain the grip detection map corresponding to the above target monitoring image.

[0109] This application calculates the IOU of bounding boxes and combines it with category probabilities to not only determine whether the mobile phone screen is correctly identified, but also evaluate its potential usage scenarios. In addition, the use of non-maximum suppression technology further optimizes the detection results, eliminates the interference of overlapping bounding boxes, and ensures the accuracy and reliability of the results.

[0110] In some optional implementations, before step S203, the following steps are further included:

[0111] Pull out the screen content training dataset and screen content test dataset from the historical image database;

[0112] Perform data preprocessing on the screen content training graph to obtain target content feature parameters;

[0113] Input the target content feature parameters into a preset general model, and train the preset general model using the screen content training dataset to obtain a content detection model;

[0114] Import the screen content test dataset into the content detection model and output the content detection results;

[0115] Based on the content detection results and the preset standard results, the back propagation algorithm is used for fitting to obtain the content loss error, and the content detection model is updated based on the content loss error.

[0116] In this embodiment, the screen content training dataset includes at least one screen content training image, and the screen content test dataset includes at least one screen content test image. Specifically, the screen content training dataset may include common office tool interfaces such as word processing software, spreadsheets, and presentation software, as well as screen images captured from various social media platforms (such as Facebook, Twitter, Instagram, etc.) and communication software (such as WhatsApp, Telegram, etc.). The screen content test dataset may be an independent image set separated from the screen content training dataset, used to evaluate the effectiveness of the content detection model.

[0117] In a possible embodiment, 1,000 images of the black screen and bright screen of the mobile phone can be collected respectively, and then 20,000 images of common objects in mobile phone photos, such as photographed documents, photographed computer pads, photographed people, photographed faces, photographed chairs, counters and other images taken casually, are collected as the above-mentioned screen content training data set and screen content test data set. During the training process, the images of the above-mentioned screen content training data set and screen content test data set are scaled to 224x224 pixels and converted into vector tensors. The data set is indexed using the DataLoader object of PyTorch (the training data index range is 0 to the training set size, and the verification data index range is 0 to the verification set size). Two V100 graphics cards (16G video memory) are used to train the MobileV3 network (i.e., the above-mentioned content detection model). The loss function is the cross entropy loss, the optimizer is Adam, the maximum training rounds are set to 50, the training samples are traversed in each round of training, the cross entropy loss between the prediction result and the label is calculated, back propagation optimization is performed, the accuracy of each round is recorded, and the optimal model weight is updated, and the batch is set. The size is 64, and the learning rate is 0.00001 and remains unchanged. Although the maximum number of epochs is 50, during actual training, it is observed that the loss is close to 0 and stable in the 6th epoch, and the accuracy reaches 0.99. Therefore, training is manually stopped, the optimal model weights are saved, and the above content detection model is updated.

[0118] This application extracts a screen content training dataset and a screen content test dataset from a historical image database, preprocesses the screen content training dataset to extract key target content feature parameters, and then uses the target content feature parameters to train a content detection model. After the content detection model is developed, its performance is evaluated on the test dataset, and the model is optimized using the backpropagation algorithm to reduce prediction errors and improve detection accuracy.

[0119] In some optional implementations, step S203 includes the following steps:

[0120] Input the grip detection image into the screen-on detection layer to perform screen-on detection and obtain the screen-on detection result;

[0121] The screen-on result is used as the input of the content recognition layer to perform content identification processing and obtain the content discrimination result.

[0122] In this embodiment, the above-mentioned screen-on detection result may include a mobile phone screen-on result and a mobile phone screen-off result.

[0123] In this embodiment, the content detection model includes a bright screen detection layer and a content recognition layer. Specifically, the bright screen detection layer is responsible for identifying the bright screen area in the image, that is, the part of the screen that displays information. The bright screen detection layer can determine which areas are active display screens by analyzing the lighting pattern and brightness distribution of the image. The bright screen is usually brighter than the surrounding environment and displays color or bright white light. The bright screen detection layer can be constructed based on the Sobel or Canny edge detection algorithm as a framework; the content recognition layer can be used to parse the specific content displayed on the screen, including text, images, icons or any other visual elements. The content recognition layer can use deep learning technologies such as convolutional neural networks (CNN) or natural language processing (NLP) algorithms to understand and classify the semantics and structure of the screen content.

[0124] In a possible embodiment, the above-mentioned holding detection image is first input into the above-mentioned bright screen detection layer, and the mobile phone screen is judged as black or bright to obtain a bright screen detection result. If the bright screen detection result is a black screen result of the mobile phone, the above-mentioned content judgment result is directly judged as safe content. If the bright screen detection result is a bright screen result of the mobile phone, the above-mentioned bright screen result is used as the input of the content recognition layer for content identification processing to detect whether the mobile phone screen contains a photo button. If so, the above-mentioned content judgment result is judged as sensitive content. If not, the above-mentioned content judgment result is judged as safe content.

[0125] This application can achieve high-precision screen content security monitoring by combining the screen light detection layer and the content recognition layer. The screen light detection layer first determines the screen state of the holding mobile phone, distinguishes between the bright screen and the black screen, and effectively identifies the active display screen, laying the foundation for further content analysis. For images in the bright screen state, the content recognition layer further analyzes the content displayed on the screen to detect whether there are sensitive elements such as the photo button, thereby accurately determining the sensitivity of the content.

[0126] In some optional implementations, step S204 includes the following steps:

[0127] Perform hand detection on the grip detection image to obtain the target hand image;

[0128] Perform joint deconstruction on the target hand graph to obtain the target joint structure graph;

[0129] The bending angle of the target joint structure diagram is detected to obtain the action discrimination result.

[0130] In this embodiment, the above-mentioned target hand image can be a specific area separated from the grip detection image. The above-mentioned specific area clearly shows the hand and its gripping state. The above-mentioned target hand image focuses on showing the shape and position of the hand, which is an accurate extraction and focusing processing of the hand area in the grip detection image.

[0131] In this embodiment, the above-mentioned target joint structure diagram can be an image further analyzed and drawn from the above-mentioned target hand diagram, which shows in detail the position and mutual connection of each joint of the hand. It can accurately depict the skeletal structure and joint layout of the hand through methods such as joint point detection and joint connection analysis, providing necessary deconstruction information for subsequent motion analysis.

[0132] In one possible embodiment, a hand detection model can be created using the MediaPipe application framework to output position information of 21 key points, including x, y, and z, where x and y are normalized coordinates, and z represents the landmark depth. The depth at the wrist is used as the origin, and the above 21 key points correspond to the joints of the palm.

[0133] Specifically, regarding the specific implementation process of performing bending angle detection on the target joint structure diagram and obtaining the action discrimination result, this application will provide further detailed descriptions in subsequent specific embodiments and will not be elaborated on here.

[0134] The hand detection model created in this application through the MediaPipe application framework can achieve high-precision hand and joint recognition. The above hand detection model identifies 21 hand key points and provides detailed information including x, y, and z coordinates, where x and y are normalized coordinates, and z represents the position relative to the depth of the wrist. It can accurately capture hand movements and the movement of each joint, enhance the performance of the gesture recognition system, enable it to accurately respond to user commands in more complex interactive environments, and greatly improve the naturalness and response speed of interaction in virtual reality and augmented reality applications.

[0135] In some optional implementations, the above step of “detecting the bending angle of the target joint structure diagram to obtain the action discrimination result” includes the following steps:

[0136] Calculate the bending angle difference of the thumb joint to obtain the thumb curvature;

[0137] Calculate the bending angle difference of the index finger joint to obtain the index finger bending degree;

[0138] The action discrimination result is determined based on the curvature of the thumb and the curvature of the thumb.

[0139] In this embodiment, the above-mentioned thumb curvature can be the angle change of the thumb from the straight state to the current state. The thumb curvature can be determined by calculating the angle difference between the two main joints of the thumb - the base joint and the interphalangeal joint. The angle difference reflects the degree of curvature of the thumb and can be used to determine the specific type of gesture or the intention of the action.

[0140] In this embodiment, the above-mentioned index finger curvature can be the angle change of the index finger from the straight state to the current state. The index finger curvature can be determined by calculating the angle difference between the two main joints of the index finger—the base joint and the interphalangeal joint. The angle difference reflects the degree of curvature of the index finger and can be used to judge the specific type of gesture or the intention of the action.

[0141] In a possible embodiment, when the thumb is used to click the screen to take a picture, the degree of bending of the thumb increases, and whether the picture was clicked is determined by judging the angle of the thumb and the difference between the previous and next angles. According to the empirical value, when the latter angle is greater than 53 degrees and the difference is greater than 5 degrees, it is determined that the thumb was used to click the screen to take a picture.

[0142] In a possible embodiment, when the screen is clicked with the index finger to take a picture, whether the picture was clicked can be determined by judging the angle between the second joint of the index finger and the third joint of the index finger and the difference between the front and back angles; if it is judged based on experience that the latter angle is greater than 60 degrees and the difference is greater than 5 degrees, it is determined that the screen was clicked with the index finger to take the picture.

[0143] This application determines whether a finger has clicked by monitoring changes in joint angles. Specifically, it determines whether the angle changes of the finger joints exceed a set threshold. This can improve the response accuracy and naturalness of the above-mentioned behavior recognition system to user gestures.

[0144] It is understandable that in the specific implementation of this application, related data such as target monitoring image sequences, target shooting behaviors, and grip detection images are involved. When the embodiments in this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0145] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned target monitoring image sequence, target shooting behavior, grip detection image and other information, the above-mentioned target monitoring image sequence, target shooting behavior, grip detection image and other information can also be stored in a blockchain node.

[0146] The blockchain referred to in this application is a new application model for computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Blockchain is essentially a decentralized database, a series of data blocks generated using cryptographic methods. Each data block contains information about a batch of network transactions, which is used to verify the validity of this information (to prevent counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, the platform product service layer, and the application service layer.

[0147] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0148] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0149] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware via computer-readable instructions. The computer-readable instructions can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes in the above-described method embodiments. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0150] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0151] Further references Figure 3 , as a response to the above Figure 2 In order to realize the method shown in FIG, the present application provides an embodiment of a behavior identification device. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0152] like Figure 3As shown, the behavior identification device 300 of this embodiment includes: an acquisition module 301, a grip detection module 302, a identification module 303, a joint detection module 304 and a determination module 305.

[0153] An acquisition module 301 is configured to acquire a target monitoring image sequence, wherein the target monitoring image sequence includes at least one target monitoring picture;

[0154] A grip detection module 302 is configured to perform grip detection processing on the target monitoring image to generate a grip detection image corresponding to each target monitoring image;

[0155] The discrimination module 303 is configured to perform content discrimination processing on the grip detection image to obtain a content discrimination result;

[0156] The joint detection module 304 is used to perform joint detection processing on the grip detection image to obtain an action discrimination result;

[0157] The determination module 305 is configured to determine the target shooting behavior based on the content identification result and the action identification result.

[0158] In a possible embodiment, before the holding detection module 302, the method further includes:

[0159] A first pulling module is configured to pull a gripping picture training dataset and a gripping picture test dataset from a historical image database, wherein the gripping picture training dataset includes at least one gripping training picture, and the gripping picture test dataset includes at least one gripping test picture;

[0160] A first preprocessing module is used to perform data preprocessing on the grip training image to obtain target grip feature parameters;

[0161] a first input module, configured to input the target grip feature parameters into a preset universal model, and train the preset universal model using the grip image training dataset to obtain a grip detection model;

[0162] A first output module, configured to import the grip image test dataset into the grip detection model and output a detection result;

[0163] The first updating module is configured to perform fitting based on the detection result and the preset standard result using a back propagation algorithm to obtain a loss error, and update the grip detection model based on the loss error.

[0164] In a possible embodiment, the grip detection module 302 includes:

[0165] A boundary prediction submodule is used to input the target monitoring image into the input layer to perform boundary prediction and obtain boundary confidence;

[0166] A category screening submodule, configured to input the boundary confidence into the prediction layer for category screening to obtain a boundary prediction result;

[0167] The image cropping submodule is used to input the boundary prediction result into the output layer for image cropping, and output the grip detection image corresponding to the target monitoring image.

[0168] In a possible embodiment, before the determination module 303, the method further includes:

[0169] A second pulling module is configured to pull a screen content training data set and a screen content test data set from the historical image database, wherein the screen content training data set includes at least one screen content training image, and the screen content test data set includes at least one screen content test image;

[0170] A second preprocessing module is used to perform data preprocessing on the screen content training image to obtain target content feature parameters;

[0171] A second input module is configured to input the target content feature parameters into a preset universal model, and train the preset universal model using the screen content training data set to obtain a content detection model;

[0172] A second output module, configured to import the screen content test dataset into the content detection model and output a content detection result;

[0173] The second updating module is used to perform fitting based on the content detection result and the preset standard result using a back propagation algorithm to obtain a content loss error, and update the content detection model based on the content loss error.

[0174] In a possible embodiment, the determination module 303 includes:

[0175] a screen-on detection submodule, configured to input the grip detection image into the screen-on detection layer to perform screen-on detection and obtain a screen-on detection result;

[0176] The content identification submodule is used to use the bright screen detection result as the input of the content recognition layer to perform content identification processing to obtain a content discrimination result.

[0177] In a possible embodiment, the joint detection module 304 includes:

[0178] A hand detection processing submodule, configured to perform hand detection processing on the grip detection image to obtain a target hand image;

[0179] A joint deconstruction submodule, configured to perform joint deconstruction on the target hand graph to obtain a target joint structure graph;

[0180] The bending angle detection submodule is used to perform bending angle detection on the target joint structure diagram to obtain the action discrimination result.

[0181] In a possible embodiment, the bending angle detection submodule includes:

[0182] a first calculation unit, configured to calculate the bending angle difference of the thumb joint to obtain the curvature of the thumb;

[0183] a second calculation unit, configured to calculate the bending angle difference of the index finger joint to obtain the curvature of the index finger;

[0184] A determination unit is used to determine the action discrimination result based on the curvature of the thumb and the curvature of the thumb.

[0185] In this embodiment, a target monitoring image sequence is obtained, holding detection processing is performed on the target monitoring pictures, and a holding detection picture corresponding to each target monitoring picture is generated. Content discrimination processing is performed on the holding detection pictures to obtain a content discrimination result. Joint detection processing is performed on the holding detection pictures to obtain an action discrimination result. Based on the content discrimination result and the action discrimination result, the target shooting behavior is determined. By real-time monitoring and analysis of the individual behavior of users within the monitoring range, and combining the specific display content of the user device, potential illegal shooting actions can be accurately identified, thereby improving the ability to prevent the leakage of sensitive information.

[0186] In this embodiment, the operations performed by the above units or modules correspond one-to-one to the steps of the behavior identification method in the above embodiment, and are not described in detail here.

[0187] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0188] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0189] The computer device may be a desktop computer, notebook computer, PDA, cloud server, etc. The computer device may interact with the user via a keyboard, mouse, remote control, touchpad, or voice control device.

[0190] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the behavior identification method. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0191] In some embodiments, the processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions or process data stored in the memory 41, such as computer-readable instructions for executing the behavior identification method.

[0192] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0193] This embodiment provides a computer device, which obtains a target monitoring image sequence, performs grip detection processing on the target monitoring images, generates a grip detection image corresponding to each target monitoring image, performs content discrimination processing on the grip detection images to obtain a content discrimination result, performs joint detection processing on the grip detection images to obtain an action discrimination result, determines the target shooting behavior based on the content discrimination result and the action discrimination result, and accurately identifies potential illegal shooting actions by real-time monitoring and analysis of the individual behavior of users within the monitoring range, and combining the specific display content of the user device, thereby improving the prevention capability against the leakage of sensitive information.

[0194] The present application also provides another embodiment, namely, providing a computer-readable storage medium, which stores computer-readable instructions. The computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the behavior judgment method as described above.

[0195] This embodiment provides a computer-readable storage medium, which obtains a target monitoring image sequence, performs grip detection processing on the target monitoring images, generates a grip detection image corresponding to each target monitoring image, performs content discrimination processing on the grip detection images to obtain a content discrimination result, performs joint detection processing on the grip detection images to obtain an action discrimination result, determines the target shooting behavior based on the content discrimination result and the action discrimination result, and accurately identifies potential illegal shooting actions by real-time monitoring and analysis of the individual behavior of users within the monitoring range, and combining the specific display content of the user device, thereby improving the prevention capability of sensitive information leakage.

[0196] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0197] Obviously, the embodiments described above are only some of the embodiments of the present application, rather than all of the embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned specific embodiments, or to make equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the present application specification and the accompanying drawings, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present application.

Claims

1. A behavior identification method, characterized in that: The steps include: Acquire a target monitoring image sequence, wherein the target monitoring image sequence includes at least one target monitoring picture; Performing grip detection processing on the target monitoring pictures to generate a grip detection picture corresponding to each target monitoring picture; Performing content discrimination processing on the holding detection image to obtain a content discrimination result; Performing joint detection processing on the grip detection image to obtain an action discrimination result; A target shooting behavior is determined based on the content identification result and the action identification result.

2. The behavior identification method according to claim 1, characterized in that: Before the step of performing grip detection processing on the target monitoring picture to generate a grip detection picture corresponding to each target monitoring picture, the step further includes: Pulling a gripping picture training data set and a gripping picture test data set from a historical image database, wherein the gripping picture training data set includes at least one gripping training picture, and the gripping picture test data set includes at least one gripping test picture; Performing data preprocessing on the grip training image to obtain target grip feature parameters; Inputting the target holding characteristic parameters into a preset universal model, and training the preset universal model using the holding picture training data set to obtain a holding detection model; Importing the holding picture test data set into the holding detection model, and outputting the detection result; Based on the detection result and the preset standard result, a back propagation algorithm is used to perform fitting to obtain a loss error, and the grip detection model is updated based on the loss error.

3. The behavior identification method according to claim 2, characterized in that: The holding detection model includes an input layer, a prediction layer, and an output layer. The step of performing holding detection processing on the target monitoring picture to generate a holding detection picture corresponding to each target monitoring picture specifically includes: Inputting the target monitoring picture into the input layer to perform boundary prediction and obtain boundary confidence; Inputting the boundary confidence into the prediction layer for category screening to obtain a boundary prediction result; The boundary prediction result is input into the output layer for image cropping, and a holding detection image corresponding to the target monitoring image is output.

4. The behavior identification method according to claim 2, characterized in that: Before the step of performing content identification processing on the holding detection image to obtain a content identification result, the method specifically includes: Pulling a screen content training data set and a screen content test data set from the historical image database, wherein the screen content training data set includes at least one screen content training picture, and the screen content test data set includes at least one screen content test picture; Performing data preprocessing on the screen content training graph to obtain target content feature parameters; Inputting the target content feature parameters into a preset universal model, and training the preset universal model using the screen content training data set to obtain a content detection model; Importing the screen content test data set into the content detection model, and outputting the content detection result; Based on the content detection result and the preset standard result, a back propagation algorithm is used for fitting to obtain a content loss error, and the content detection model is updated based on the content loss error.

5. The behavior identification method according to claim 4, characterized in that: The content detection model includes a bright screen detection layer and a content recognition layer. The step of performing content discrimination processing on the holding detection picture to obtain a content discrimination result specifically includes: Inputting the holding detection picture into the bright screen detection layer to perform bright screen detection, and obtaining a bright screen detection result; The bright screen detection result is used as the input of the content recognition layer to perform content identification processing to obtain a content discrimination result.

6. The behavior identification method according to claim 1, characterized in that: The step of performing joint detection processing on the holding detection image to obtain an action identification result also includes: Performing hand detection processing on the holding detection picture to obtain a target hand picture; Performing joint deconstruction on the target hand graph to obtain a target joint structure graph; The bending angle of the target joint structure diagram is detected to obtain the action discrimination result.

7. The behavior identification method according to claim 6, characterized in that: The target joint structure diagram includes a thumb joint and an index finger joint, and the step of performing bending angle detection on the target joint structure diagram to obtain the action discrimination result specifically includes: Calculating the bending angle difference of the thumb joint to obtain the curvature of the thumb; Calculating the bending angle difference of the index finger joint to obtain the index finger bending degree; The action discrimination result is determined based on the thumb curvature and the thumb bending degree.

8. A behavior identification device, characterized in that: include: An acquisition module, used to acquire a target monitoring image sequence, wherein the target monitoring image sequence includes at least one target monitoring picture; A holding detection module, used to perform holding detection processing on the target monitoring picture to generate a holding detection picture corresponding to each target monitoring picture; A discrimination module, used for performing content discrimination processing on the holding detection image to obtain a content discrimination result; A joint detection module, used to perform joint detection processing on the holding detection image to obtain an action discrimination result; The determination module is used to determine the target shooting behavior based on the content discrimination result and the action discrimination result.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the behavior identification method according to any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the behavior identification method according to any one of claims 1 to 7 are implemented.

Images