Personnel invasion visual identification method, device, equipment, medium and program

Abstract

The invention relates to a personnel limit invasion visual identification method, device and equipment, a storage medium and a computer program, and the method comprises the steps: obtaining monitoring video data in a preset monitoring range, carrying out the denoising of the monitoring video data, obtaining the denoised video data, extracting the video features of the denoised video data, and obtaining the video features of the denoised video data; performing multi-layer dissimilatory convolution processing on the video features by using a pre-trained personnel invasion visual identification model to obtain personnel behavior features in the monitoring video data, and judging whether a personnel invasion event exists in the monitoring video data according to the personnel behavior features, and if the personnel invasion event exists in the monitoring video data, sending alarm information to preset personnel. According to the invention, the efficiency of identifying the personnel invasion event through video monitoring is improved, and if the personnel invasion event is found, the alarm information is sent to the police in time.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular to a method, device, equipment, storage medium and computer program for visually identifying personnel intrusion. Background Art

[0002] With the rapid development of video surveillance technology, high-definition, high-frame-rate and real-time transmitted video streams enable us to capture and analyze more detailed information. Surveillance videos provide a rich data source for behavior recognition. The videos of the monitored areas collected in real time by cameras and other devices contain a large number of behavior patterns and dynamic changes.

[0003] In practical applications, for example, in public transportation places such as train platforms, monitoring systems play a vital role in order to strictly comply with safety regulations and prevent pedestrians from entering dangerous areas. These monitoring systems help police officers to detect abnormal situations in time and take necessary measures to ensure the safety of passengers and staff through real-time monitoring and video playback. However, the efficiency of video monitoring in identifying intrusion incidents is low in the existing technology, and there is a problem that alarm information cannot be sent to police officers in time. Summary of the invention

[0004] The present application provides a method, device, equipment, storage medium and computer program for visually identifying people intruding from a restricted area, so as to solve the problem that video surveillance is inefficient in identifying people intruding from a restricted area and cannot send alarm information to police officers in a timely manner.

[0005] In a first aspect, the present application provides a method for visually identifying a person who has intruded into a restricted area, comprising:

[0006] Acquire surveillance video data within a preset surveillance range, and denoise the surveillance video data to obtain denoised video data;

[0007] Extracting video features of the denoised video data;

[0008] Using a pre-trained personnel intrusion visual recognition model to perform multi-layer alienation convolution processing on the video features, to obtain personnel behavior features in the monitoring video data;

[0009] Determine whether there is a person intrusion event in the surveillance video data according to the person's behavior characteristics;

[0010] If there is a human intrusion event in the monitoring video data, an alarm message is sent to a preset person.

[0011] In some embodiments, denoising the surveillance video data to obtain denoised video data includes:

[0012] Decomposing each frame of the surveillance video data into wavelet coefficients of different frequencies by using wavelet transform;

[0013] Performing inverse transformation on wavelet coefficients greater than a preset coefficient threshold to obtain transformed video data;

[0014] Counting the number of pixels of a preset gray level in each frame of the transformed video data;

[0015] Generating a grayscale histogram using the statistical result of the number of pixels at the preset grayscale level;

[0016] Calculating the cumulative distribution function of the grayscale histogram;

[0017] Normalizing the cumulative distribution function to obtain a normalized function;

[0018] Calculating the updated gray level of the preset gray level pixel using the normalization function;

[0019] Mapping the updated grayscale level to each frame image in the transformed video data to obtain an updated frame image;

[0020] All updated frame images are concatenated to obtain denoised video data.

[0021] In some embodiments, extracting video features of the denoised video data includes:

[0022] Extracting key frames from the denoised video data;

[0023] Selecting one of the key frames as a target frame one by one, and converting the target frame into a target matrix;

[0024] Performing a two-dimensional convolution on the target matrix to obtain a convolution matrix;

[0025] Performing global average pooling on the convolution matrix to obtain a pooling matrix;

[0026] The video features of the denoised video data are generated according to the pooling matrix.

[0027] In some embodiments, the use of a pre-trained personnel intrusion visual recognition model to perform multi-layer alienation convolution processing on the video features to obtain personnel behavior features in the surveillance video data includes:

[0028] Converting the video features into an initial feature matrix;

[0029] Using the two-dimensional convolution LSTM layer in the personnel intrusion visual recognition model to perform temporal convolution on the initial feature matrix to obtain a first convolution matrix;

[0030] Performing three-dimensional convolution on the first convolution matrix using the three-dimensional convolution layer in the person intrusion visual recognition model to obtain a second convolution matrix;

[0031] Pooling the second convolution matrix to obtain a pooling matrix;

[0032] The pooling matrix is ​​fully connected to obtain the behavior characteristics of the personnel in the monitoring video data.

[0033] In some embodiments, judging whether there is a person intrusion event in the surveillance video data according to the person behavior characteristics includes:

[0034] Calculate the probability of the personnel behavior characteristics matching multiple preset personnel intrusion events;

[0035] When the probability of the human behavior characteristics matching any human intrusion event is greater than a preset probability threshold, it is determined that there is a human intrusion event in the surveillance video data;

[0036] When the probability of the human behavior characteristics matching all human intrusion events is less than or equal to the probability threshold, it is determined that there is no human intrusion event in the monitoring video data.

[0037] In some embodiments, sending the alarm information to a preset person includes:

[0038] Generate a key video frame according to the monitoring video data, and send the key video frame as alarm information to the preset person in at least one of the following ways:

[0039] Sending alarm information to the preset personnel via SMS;

[0040] Sending alarm information to the preset personnel via email;

[0041] The alarm information is sent to the preset personnel through the preset APP push.

[0042] In a second aspect, the present application provides a person intrusion visual recognition device, comprising:

[0043] A data denoising module is used to obtain surveillance video data within a preset surveillance range, denoise the surveillance video data, and obtain denoised video data;

[0044] A feature extraction module, used to extract video features of the denoised video data;

[0045] A data analysis module, used to perform multi-layer alienation convolution processing on the video features using a pre-trained personnel intrusion visual recognition model to obtain personnel behavior features in the monitoring video data;

[0046] A feature judgment module is used to judge whether there is a person intrusion event in the monitoring video data according to the person's behavior characteristics;

[0047] The alarm information sending module is used to send an alarm message to a preset person if there is a person intrusion event in the monitoring video data.

[0048] In a third aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the above aspects.

[0049] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method described in the above aspects when executed by a processor.

[0050] In a fifth aspect, the present application provides a computer program product, including a computer program / instructions, which implements the steps of the method described in the above aspects when the computer program is executed by a processor.

[0051] The present application provides a method, device, equipment, storage medium and computer program for visual identification of intrusion of persons, which obtains surveillance video data within a preset surveillance range, denoises the surveillance video data to obtain denoised video data, extracts video features of the denoised video data, uses a pre-trained visual identification model for intrusion of persons to perform multi-layer alienated convolution processing on the video features, obtains the behavioral features of persons in the surveillance video data, determines whether there is an intrusion of persons in the surveillance video data based on the behavioral features of the persons, and sends an alarm message to a preset person if there is an intrusion of persons in the surveillance video data. This solves the problem that the efficiency of video surveillance in identifying intrusion of persons is low and the alarm message cannot be sent to police officers in a timely manner.

[0052] 1. Use wavelet transform to denoise surveillance video data, retain useful image information, and reduce the impact of noise, effectively improving the quality of surveillance video data and the accuracy of subsequent identification of personnel intrusion events.

[0053] 2. The person intrusion visual recognition model based on convolutional long short-term memory neural network defines a two-dimensional convolutional LSTM layer to process the time series features of video data and a three-dimensional convolutional layer to convert the extracted time series features into a second convolution matrix. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The present application will be described in more detail below based on embodiments and with reference to the accompanying drawings:

[0055] Figure 1A flowchart of a method for visually identifying a person intruding into a restricted area provided in an embodiment of the present application;

[0056] Figure 2 A schematic diagram of the functional modules of a person intrusion visual recognition device provided in an embodiment of the present application;

[0057] Figure 3 A schematic diagram of the structure of an electronic device for visually identifying human intrusion provided in an embodiment of the present application.

[0058] In the drawings, the same reference numerals are used for the same components, and the drawings are not drawn to scale. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the technical solution of the present application, and to fully understand and implement how the present application applies technical means to solve technical problems and achieve the corresponding technical effects, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only embodiments of a part of the present application, not all embodiments. The embodiments of the present application and the various features in the embodiments can be combined with each other without conflict, and the technical solutions formed are all within the scope of protection of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work should belong to the scope of protection of the present application.

[0060] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0061] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0062] Based on the above background, this scheme proposes a human intrusion visual recognition scheme, which can use a pre-trained human intrusion visual recognition model to perform multi-layer alienation convolution processing on the video features to obtain the human behavior characteristics in the monitoring video data, and judge whether there is a human intrusion event in the monitoring video data based on the human behavior characteristics. Possible human intrusion events can be monitored and identified in real time. If there is a human intrusion event in the monitoring video data, an alarm message is sent in time to improve monitoring efficiency.

[0063] Embodiment 1

[0064] Figure 1 A flowchart of a method for visually identifying a person who has intruded into a restricted area provided in an embodiment of the present application is shown below. Figure 1 As shown, a method for visually identifying a person intruding into a limit, comprising:

[0065] S1. Acquire surveillance video data within a preset surveillance range, and denoise the surveillance video data to obtain denoised video data.

[0066] In an embodiment of the present invention, video data of the monitored area is collected in real time through a camera device with configured parameters, and the configuration parameters include resolution, frame rate, night vision capability, etc. The monitored video data contains information such as behavior patterns and dynamic changes of people within the range that needs to be monitored.

[0067] In the embodiment of the present invention, the step of denoising the surveillance video data to obtain denoised video data includes:

[0068] Decomposing each frame of the surveillance video data into wavelet coefficients of different frequencies by using wavelet transform;

[0069] Performing inverse transformation on wavelet coefficients greater than a preset coefficient threshold to obtain transformed video data;

[0070] Counting the number of pixels of a preset gray level in each frame of the transformed video data;

[0071] Generating a grayscale histogram using the statistical result of the number of pixels at the preset grayscale level;

[0072] Calculating the cumulative distribution function of the grayscale histogram;

[0073] Normalizing the cumulative distribution function to obtain a normalized function;

[0074] Calculating the updated gray level of the preset gray level pixel using the normalization function;

[0075] Mapping the updated grayscale level to each frame image in the transformed video data to obtain an updated frame image;

[0076] All updated frame images are concatenated to obtain denoised video data.

[0077] Wherein, each frame of the monitoring video data is decomposed into wavelet coefficients of different frequencies by using wavelet transform. The more the decomposed frequencies are, the richer the level of details that the wavelet transform can capture from the monitoring video data is.

[0078] Furthermore, the wavelet coefficients greater than the preset coefficient threshold are inversely transformed to reconstruct the denoised surveillance video frame and obtain the transformed video data. The wavelet transform will retain useful image information while reducing the impact of noise on the surveillance video data, effectively improving the quality of the surveillance video data and the accuracy of subsequent analysis.

[0079] In detail, the cumulative distribution function represents the cumulative percentage of pixels at each gray level, and the cumulative distribution function can be calculated using the following formula:

[0080]

[0081] Among them, CDF(a) represents the cumulative percentage of pixels of all gray levels with gray level less than or equal to a, P(b) represents the probability of pixels of gray level b, that is, the proportion of pixels of gray level b to the total number of pixels, and a represents the gray level a.

[0082] The cumulative distribution function can be normalized by the following formula:

[0083]

[0084] Among them, CDF(i) is the cumulative distribution function value at gray level a, CDF min is the minimum value of the cumulative distribution function, which is usually zero during processing, H is the total number of gray levels, and a indicates that the gray level is a.

[0085] Histogram equalization redistributes the grayscale distribution of an image to make the grayscale distribution of pixels in the image more uniform, thereby improving the visual quality and detail visibility of the image.

[0086] Acquiring surveillance video data within a preset surveillance range and configuring camera parameters is conducive to optimizing the video acquisition effect, cleaning the surveillance video data, and improving the contrast and clarity of the video in order to better extract features.

[0087] S2. Extracting video features of the denoised video data.

[0088] In the embodiment of the present invention, extracting the video features of the denoised video data includes:

[0089] Extracting key frames from the denoised video data;

[0090] Selecting one of the key frames as a target frame one by one, and converting the target frame into a target matrix;

[0091] Performing a two-dimensional convolution on the target matrix to obtain a convolution matrix;

[0092] Performing global average pooling on the convolution matrix to obtain a pooling matrix;

[0093] The video features of the denoised video data are generated according to the pooling matrix.

[0094] In the embodiment of the present invention, the two-dimensional convolution operation extracts video features, such as edge features, feature enhancement, etc., by sliding a convolution kernel on the two-dimensional plane of the target matrix. Global average pooling compresses the entire convolution matrix into one value, which can effectively reduce the number of parameters and improve the efficiency of subsequent image analysis processing.

[0095] In detail, the video features include edge features, corner points, SIFT features, and gradient histograms of each frame of the cleaning video.

[0096] The step of generating edge features includes: using the Sobel operator to calculate the gradient strength and gradient direction of each frame of the denoised video data to obtain a gradient image; screening out a number of strong edge pixels whose gradient values ​​of the gradient image are greater than a preset gradient threshold; connecting the several strong edge pixels to obtain edge features of each frame of the denoised video data. The Canny edge detection algorithm can effectively extract edge information from each frame of the denoised video data, and while maintaining edge details, suppress noise and non-edge information, which is conducive to obtaining clear and accurate edge lines.

[0097] Harris corner detector is a corner detection algorithm that determines corners by calculating the eigenvalues ​​of the pixel grayscale matrix of each frame of the denoised video data. Harris corner detector has low computational complexity and can very accurately locate the corners in each frame of the denoised video data, which is beneficial for subsequent image registration and target tracking.

[0098] The SIFT algorithm is an image feature descriptor that is stable at different scales and rotations. The SIFT algorithm extracts local feature points from each frame of the denoised video data and generates a vector that can describe the local feature points. The vector describing the local feature points is the SIFT feature, which is invariant to image scaling, rotation, and partial occlusion, and can achieve high-level image matching and object recognition, and is robust and accurate for target detection and image registration in complex scenes.

[0099] The steps of generating HOG features include: converting each frame of the denoised video data into a grayscale cleaned image; calculating the gradient amplitude and gradient direction of each pixel in each grayscale cleaned image; dividing each grayscale cleaned image into a number of cells; weighting the gradient amplitude in the gradient direction of each pixel in the cell to obtain a weighted histogram; normalizing the weighted histogram to obtain a gradient direction histogram; and connecting all gradient direction histograms in series to obtain HOG features. HOG features are an effective description of the local structure and texture features of an image. By calculating the gradient direction and amplitude of an image, HOG can capture important information about the edge and contour of an object, and has a certain robustness to changes in illumination.

[0100] Extracting the video features of the denoised video data is conducive to improving the comprehensive description and analysis capabilities of images or targets. Among them, edge features can capture the contour and boundary information of objects, corner features help detect significant structural changes in images, SIFT features provide a relatively robust local description of scale and rotation changes, and HOG features can effectively characterize the texture and local structure of images. The comprehensive use of these features can improve the accuracy and robustness of visual recognition of personnel intrusion.

[0101] S3. Use a pre-trained personnel intrusion visual recognition model to perform multi-layer alienation convolution processing on the video features to obtain personnel behavior characteristics in the monitoring video data.

[0102] In the embodiment of the present invention, the multi-layer alienation convolution processing is performed on the video features using the pre-trained personnel intrusion visual recognition model to obtain the personnel behavior features in the monitoring video data, including:

[0103] Converting the video features into an initial feature matrix;

[0104] Using the two-dimensional convolution LSTM layer in the personnel intrusion visual recognition model to perform temporal convolution on the initial feature matrix to obtain a first convolution matrix;

[0105] Performing three-dimensional convolution on the first convolution matrix using the three-dimensional convolution layer in the person intrusion visual recognition model to obtain a second convolution matrix;

[0106] Pooling the second convolution matrix to obtain a pooling matrix;

[0107] The pooling matrix is ​​fully connected to obtain the behavior characteristics of the personnel in the monitoring video data.

[0108] In detail, the two-dimensional convolutional LSTM layer mainly includes four parts: input gate, forget gate, output gate and cell state update, among which the update of feature information in the input gate and forget gate affects the transformation of the cell state.

[0109] Furthermore, when the initial feature matrix enters the two-dimensional convolutional LSTM layer, the input gate selectively uses the initial feature matrix to update the cell state, and fuses the new initial feature matrix information into the cell state so as to capture the important information of the initial feature matrix.

[0110] The forget gate needs to decide which part of the historical feature matrix left by historical recognition in the cell state at the moment before the initial feature matrix enters the two-dimensional convolutional LSTM layer needs to be cleared, and the forget gate updates the long-term memory by controlling the degree of forgetting.

[0111] The cell state update is responsible for calculating and updating the cell state of the LSTM unit in the two-dimensional convolutional LSTM layer. It combines the initial feature matrix information to be added by the input gate and the historical feature matrix information to be removed by the forget gate to update the cell state of the current time step.

[0112] When the initial feature matrix enters the two-dimensional convolutional LSTM layer, the output gate determines how the cell state at the time of entry is passed to the next time step or as the final output. The output gate generates the final output based on the input and memory state at the current moment or passes it to the next three-dimensional convolutional layer.

[0113] In detail, the person intrusion visual recognition model based on convolutional long short-term memory neural network defines a two-dimensional convolutional LSTM layer for processing the time series features of video data and a three-dimensional convolutional layer for converting the extracted time series features into a second convolution matrix.

[0114] The two-dimensional convolution LSTM layer in the person intrusion visual recognition model is used to perform time series convolution on the initial feature matrix to obtain a first convolution matrix, where the first convolution matrix refers to the time series features of the video features. The three-dimensional convolution layer in the person intrusion visual recognition model is used to perform three-dimensional convolution on the first convolution matrix to obtain a second convolution matrix, where the second convolution features refer to the convolution matrix obtained by convolving the time series features of the video features.

[0115] Among them, the loss function in the pre-trained person intrusion visual recognition model is:

[0116]

[0117] Wherein, L represents the loss function, N represents the total number of training data of the personnel behavior characteristics in the monitoring video data, C represents the total number of behavior characteristic categories, and yij Indicates the true label of the i-th training data belonging to the j-th category behavior feature, It represents the personnel behavior characteristics predicted by the personnel intrusion visual recognition model.

[0118] The loss function calculates the weighted logarithmic probability sum of the number of training data for all personnel behavior characteristics and the number of all behavior characteristic categories. The model parameters are optimized through the gradient descent algorithm to minimize the loss function.

[0119] Specifically, the gradient of the loss function L with respect to the parameters of the person intrusion visual recognition model is calculated. The gradient is the partial derivative of the loss function with respect to each parameter, which represents the rate of change of the loss function in the parameter space. The parameters of the person intrusion visual recognition model are updated according to the calculated gradient, so that the loss function L is gradually reduced. The formula for updating the parameters of the person intrusion visual recognition model is:

[0120]

[0121] Among them, θ represents the parameters of the human intrusion visual recognition model, α represents the learning rate, Represents the gradient of the loss function with respect to the parameter θ. Iteratively execute the steps of calculating the gradient of the loss function L with respect to the parameters of the person intrusion visual recognition model and updating the parameters of the person intrusion visual recognition model according to the calculated gradient until the loss function converges to a smaller value or reaches a preset number of iterations.

[0122] Optimizing model parameters through the gradient descent algorithm can effectively help the human intrusion visual recognition model find the optimal parameter combination that minimizes the loss function. It can not only accelerate model convergence and improve training efficiency, but also cope with complex nonlinear relationships and large-scale data sets, making the model more robust and reliable in practical applications.

[0123] By analyzing the video features using a pre-trained human intrusion visual recognition model, accurate identification and prediction of human intrusion incidents can be achieved by effectively capturing and analyzing the features in the surveillance video data, thereby improving the response speed and accuracy of the security monitoring system and effectively reducing the false alarm rate.

[0124] S4. Determine whether there is any human intrusion event in the surveillance video data based on the human behavior characteristics.

[0125] In the embodiment of the present invention, judging whether there is a person intrusion event in the monitoring video data according to the person behavior characteristics includes:

[0126] Calculate the probability of the personnel behavior characteristics matching multiple preset personnel intrusion events;

[0127] When the probability of the human behavior characteristics matching any human intrusion event is greater than a preset probability threshold, it is determined that there is a human intrusion event in the surveillance video data;

[0128] When the probability of the human behavior characteristics matching all human intrusion events is less than or equal to the probability threshold, it is determined that there is no human intrusion event in the monitoring video data.

[0129] In detail, the distance values ​​between the personnel behavior characteristics and multiple preset personnel intrusion events can be calculated by Euclidean distance algorithm, cosine distance algorithm, etc., and the calculated distance values ​​are normalized to obtain the probability of matching the personnel behavior characteristics with multiple preset personnel intrusion events.

[0130] In the embodiment of the present invention, after determining that there is no human intrusion event in the surveillance video data, the process may return to S1 of the step of obtaining surveillance video data within a preset surveillance range.

[0131] S5. If there is a human intrusion event in the monitoring video data, an alarm message is sent to a preset person.

[0132] In the embodiment of the present invention, sending the alarm information to the preset personnel includes:

[0133] Generate a key video frame according to the monitoring video data, and send the key video frame as alarm information to the preset person in at least one of the following ways:

[0134] Sending alarm information to the preset personnel via SMS;

[0135] Sending alarm information to the preset personnel via email;

[0136] The alarm information is sent to the preset personnel through the preset APP push.

[0137] In detail, the motion detection technology is used to detect the behavior of the personnel in each frame of the monitoring video data to generate a key video frame. Among them, the monitoring video image of each frame in the monitoring video data is compared with the reference frame, and the difference between the pixels of a frame in the monitoring video data and the reference frame is calculated. When the difference exceeds the preset difference threshold, the monitoring video image frame compared with the reference frame at this time will be marked as a key video frame. The key video frame is sent to the preset personnel as an alarm message through one of the methods: such as SMS, email, APP push, etc.

[0138] If there is a person intruding into the restricted area in the monitoring video data, an alarm message will be sent to the preset personnel, which can monitor and identify possible person intruding into the restricted area in real time, send alarm messages in time, effectively prevent and reduce potential security threats and losses, and protect the safety of personnel.

[0139] Embodiment 2

[0140] like Figure 2 As shown, it is a functional module diagram of a person intrusion visual recognition device 100 provided for this embodiment.

[0141] The person-intrusion visual recognition device 100 of the present invention can be installed in an electronic device. According to the functions to be implemented, the person-intrusion visual recognition device 100 may include a data denoising module 101, a feature extraction module 102, a data analysis module 103, a feature judgment module 104, and an alarm information sending module 105. The module of the present invention may also be referred to as a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, and is stored in the memory of the electronic device.

[0142] In this embodiment, the functions of each module / unit are as follows:

[0143] The data denoising module 101 is used to obtain surveillance video data within a preset surveillance range, denoise the surveillance video data, and obtain denoised video data;

[0144] A feature extraction module 102, used to extract video features of the denoised video data;

[0145] The data analysis module 103 is used to perform multi-layer alienation convolution processing on the video features using a pre-trained personnel intrusion visual recognition model to obtain personnel behavior features in the monitoring video data;

[0146] A feature judgment module 104 is used to judge whether there is a person intrusion event in the monitoring video data according to the person's behavior characteristics;

[0147] The alarm information sending module 105 is used to send an alarm message to a preset person if there is a person intrusion event in the monitoring video data.

[0148] Embodiment 3

[0149] Figure 3 A schematic diagram of the structure of an electronic device for visually identifying human intrusion provided in an embodiment of the present application.

[0150] On the basis of the above embodiments, this embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the above embodiments.

[0151] In some implementations of this embodiment, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the computer program is executed by a processor, the steps of the method described in the above embodiment are implemented.

[0152] In some implementations of this embodiment, a computer program product is provided, including a computer program / instructions, characterized in that when the computer program is executed by a processor, the steps of the method described in the above embodiment are implemented.

[0153] The processor may include, but is not limited to, one or more processors or microprocessors. Each processor may be an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller, a microprocessor or other electronic components to execute the method in the above embodiments.

[0154] The computer-readable storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, and the computer-readable storage medium may include but is not limited to, for example, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, computer storage media (e.g., hard disk, floppy disk, solid-state drive, removable disk, CD-ROM, DVD-ROM, Blu-ray disc, etc.).

[0155] The computer-readable storage medium may also store at least one computer executable program / instruction, which may be, for example, a computer-readable instruction. The computer-readable storage medium includes, but is not limited to, for example, a volatile memory and / or a non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The computer-readable storage medium may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device runs the computer-readable instructions stored on the computer-readable storage medium, the various methods described above may be performed.

[0156] In addition, the computer device may also include (but not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (eg, keyboard, mouse, speaker, etc.), etc.

[0157] The processor may communicate with a communication interface of an external device via an I / O bus via a wired or wireless network.

[0158] In one embodiment, the at least one computer executable instruction may also be compiled into or constitute a software product / computer program product, wherein one or more computer executable instructions are executed by a processor to perform the various functions and / or method steps in the embodiments described in the present technology.

[0159] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the above-mentioned module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based device that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0160] It should be noted that, in this application, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element limited by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0161] Although the implementation methods disclosed in this application are as above, the above contents are only the implementation methods adopted for facilitating the understanding of this application, and are not intended to limit this application. Any technician in the technical field to which this application belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application, but the scope of patent protection of this application shall still be based on the scope defined in the attached claims.

Claims

1. A method for visually identifying a person who has intruded into a restricted area, characterized in that: include: Acquire surveillance video data within a preset surveillance range, and denoise the surveillance video data to obtain denoised video data; Extracting video features of the denoised video data; Using a pre-trained personnel intrusion visual recognition model to perform multi-layer alienation convolution processing on the video features, to obtain personnel behavior features in the monitoring video data; Determine whether there is a person intrusion event in the surveillance video data according to the person's behavior characteristics; If there is a human intrusion event in the monitoring video data, an alarm message is sent to a preset person.

2. The method according to claim 1, characterized in that The step of denoising the surveillance video data to obtain denoised video data includes: Decomposing each frame of the surveillance video data into wavelet coefficients of different frequencies by using wavelet transform; Performing inverse transformation on wavelet coefficients greater than a preset coefficient threshold to obtain transformed video data; Counting the number of pixels of a preset gray level in each frame of the transformed video data; Generating a grayscale histogram using the statistical result of the number of pixels at the preset grayscale level; Calculating the cumulative distribution function of the grayscale histogram; Normalizing the cumulative distribution function to obtain a normalized function; Calculating the updated gray level of the preset gray level pixel using the normalization function; Mapping the updated grayscale level to each frame image in the transformed video data to obtain an updated frame image; All updated frame images are concatenated to obtain denoised video data.

3. The method according to claim 1, characterized in that The extracting the video features of the denoised video data comprises: Extracting key frames from the denoised video data; Selecting one of the key frames as a target frame one by one, and converting the target frame into a target matrix; Performing a two-dimensional convolution on the target matrix to obtain a convolution matrix; Performing global average pooling on the convolution matrix to obtain a pooling matrix; The video features of the denoised video data are generated according to the pooling matrix.

4. The method according to claim 1, characterized in that The pre-trained personnel intrusion visual recognition model is used to perform multi-layer alienation convolution processing on the video features to obtain the personnel behavior features in the monitoring video data, including: Converting the video features into an initial feature matrix; Using the two-dimensional convolution LSTM layer in the personnel intrusion visual recognition model to perform temporal convolution on the initial feature matrix to obtain a first convolution matrix; Performing three-dimensional convolution on the first convolution matrix using the three-dimensional convolution layer in the person intrusion visual recognition model to obtain a second convolution matrix; Pooling the second convolution matrix to obtain a pooling matrix; The pooling matrix is ​​fully connected to obtain the behavior characteristics of the personnel in the monitoring video data.

5. The method according to claim 1, characterized in that The determining whether there is a person intrusion event in the surveillance video data according to the person behavior characteristics includes: Calculate the probability of the personnel behavior characteristics matching multiple preset personnel intrusion events; When the probability of the human behavior characteristics matching any human intrusion event is greater than a preset probability threshold, it is determined that there is a human intrusion event in the surveillance video data; When the probability of the human behavior characteristics matching all human intrusion events is less than or equal to the probability threshold, it is determined that there is no human intrusion event in the monitoring video data.

6. The method according to claim 5, characterized in that The sending of alarm information to a preset person includes: Generate a key video frame according to the monitoring video data, and send the key video frame as alarm information to the preset person in at least one of the following ways: Sending alarm information to the preset personnel via SMS; Sending alarm information to the preset personnel via email; The alarm information is sent to the preset personnel through the preset APP push.

7. A visual identification device for personnel intrusion, characterized in that: The device comprises: A data denoising module is used to obtain surveillance video data within a preset surveillance range, denoise the surveillance video data, and obtain denoised video data; A feature extraction module, used to extract video features of the denoised video data; A data analysis module, used to perform multi-layer alienation convolution processing on the video features using a pre-trained personnel intrusion visual recognition model to obtain personnel behavior features in the monitoring video data; A feature judgment module is used to judge whether there is a person intrusion event in the monitoring video data according to the person's behavior characteristics; The alarm information sending module is used to send an alarm message to a preset person if there is a person intrusion event in the monitoring video data.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images