A computer-based intelligent security monitoring platform and system

Abstract

The present invention relates to the field of intelligent security monitoring, and specifically discloses an intelligent security monitoring platform based on a computer and a system thereof, comprising: an information collection unit collects security event types and dangerous event types as well as real-time monitoring data acquired by multiple cameras; an information processing unit screens and acquires abnormal behavior information according to pre-processed security event types and dangerous event types and marks them; abnormal behavior information is identified for multiple cameras respectively; a network construction unit determines a target camera and constructs a target camera network; a security monitoring unit acquires a real-time monitoring data set of all target cameras in the target camera network within a time period; an information analysis unit performs target detection analysis on the real-time monitoring data set acquired by multiple target cameras, and determines whether there is an abnormal target individual in a target frame: if so, characteristic coding sequences of the abnormal target individuals under multiple different cameras are acquired; if not, security monitoring is continued.

Description

Technical Field

[0001] The present invention relates to the field of intelligent security monitoring, and in particular to a computer-based intelligent security monitoring platform and a system thereof. Background Art

[0002] Computer intelligent security monitoring systems are applicable to a variety of occasions, and are often used in urban security, traffic management, smart construction sites, tourist attractions, logistics warehousing, smart homes, etc. In these scenarios, security monitoring systems can help improve management efficiency, strengthen security precautions, optimize resource allocation, and provide rapid response to emergencies.

[0003] Common intelligent security monitoring systems mainly rely on intelligent video analysis technology. By introducing advanced technology, it can realize automatic recognition, tracking, classification and behavior analysis of targets in video images, providing smart security systems with more intelligent, accurate and efficient monitoring and security prevention methods. Intelligent video analysis technology has also received widespread attention and application due to its advantages in face recognition, behavior analysis, target detection, etc.

[0004] However, the existing computer intelligent security monitoring has the problem of poor synchronization between the camera risk capture status and the camera's monitoring image recognition risk information, and the construction of the camera network has the problems of inadequate risk capture, incomplete risk range feedback and untimely receipt of risk information. Summary of the invention

[0005] The purpose of the present invention is to provide a computer-based intelligent security monitoring platform and system thereof to solve the following technical problems:

[0006] How to improve the accuracy of camera risk capture and achieve synchronous recognition of abnormal target information in surveillance images by target cameras.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A computer-based intelligent security monitoring system, comprising:

[0009] An information collection unit, used to collect the types of safety events and dangerous events in the historical database for pre-processing; and to collect real-time monitoring data obtained by multiple cameras;

[0010] An information processing unit, used to filter, obtain and mark abnormal behavior information according to the pre-processed security event type and dangerous event type; and to identify abnormal behavior information for multiple cameras respectively;

[0011] A network construction unit, used to determine the camera that has identified abnormal behavior information as a target camera and build a target camera network;

[0012] A security monitoring unit is used to obtain a collection of real-time monitoring data of all target cameras in a target camera network over a period of time;

[0013] The information analysis unit is used to perform target detection analysis and target frame annotation on the real-time monitoring data set obtained by multiple target cameras, and obtain multiple image frames containing target frame annotations, and determine whether there are abnormal target individuals in the target frame according to target detection:

[0014] If so, obtain the characteristic coding sequence of the abnormal target individual under multiple different cameras;

[0015] If not, continue with security monitoring;

[0016] The information fusion unit is used to perform multimodal fusion matching on the feature coding sequence of the abnormal target individual based on the feature matching algorithm, obtain the matched camera target trajectory fragment, construct the movement trajectory information of the abnormal target individual, and update it to the security monitoring terminal in real time.

[0017] Preferably, the method by which the information processing unit screens and obtains abnormal behavior information and marks it according to the pre-processed security event type and dangerous event type is:

[0018] Obtain samples of security event types and dangerous event types in security monitoring based on big data;

[0019] Extracting three-dimensional data sets of key points from image information of safety event type samples and dangerous event type samples respectively;

[0020] The three-dimensional data set of safety event type samples is collected as positive samples, and the three-dimensional data set of dangerous event type samples is collected as negative samples and input into the convolutional neural network model for learning to obtain a trajectory recognition model;

[0021] The three-dimensional data set corresponding to the real-time monitoring data obtained by multiple cameras is input into the trajectory recognition model to judge abnormal behavior, obtain abnormal behavior information in security monitoring and mark it.

[0022] Preferably, the method of inputting the three-dimensional data set corresponding to the real-time monitoring data acquired by the camera into the trajectory recognition model for abnormal behavior judgment is:

[0023] By formula Calculate the recognition coefficient of the i-th camera ;

[0024] in, is the three-dimensional data function of the i-th camera; is the k-th 3D data function of the i-th camera Axis coordinates; is the k-th 3D data function of the i-th camera Axis coordinates; is the k-th 3D data function of the i-th camera Axis coordinates; is the standard three-dimensional data function of the i-th camera; is the standard 3D data function of the i-th camera Axis coordinates; is the standard 3D data function of the i-th camera Axis coordinates; is the standard 3D data function of the i-th camera Axis coordinates; is the average 3D data function of historical cameras; is the average 3D data function of the historical camera Axis coordinates; is the average 3D data function of the historical camera Axis coordinates; is the average 3D data function of the historical camera Axis coordinates; is the number of three-dimensional data function terms, and .

[0025] Preferably, the abnormal behavior information recognition method for multiple cameras is to use the recognition coefficient of the i-th camera The threshold interval of the standard recognition coefficient To compare:

[0026] like > , then it is judged that the current running trajectory of the i-th camera has the risk of abnormal behavior information, and the camera information is collected;

[0027] like ∈ , then it is judged that the risk of abnormal behavior information of the current running trajectory of the i-th camera is low;

[0028] like < , then it is judged that there is no abnormal behavior information in the current running trajectory of the i-th camera.

[0029] Preferably, the target detection analysis method of the information analysis unit is:

[0030] By formula Calculate the monitoring coefficient of the i-th target camera ;

[0031] in, is the total number of individual target features detected by the target camera, and ∈ ; is the jth target feature distance parameter of the i-th target camera individual; is the standard distance parameter of the jth target feature of the i-th target camera individual; is the center distance parameter of the jth target feature of the individual; is the preset weight coefficient of the jth target feature of the i-th target camera individual.

[0032] Preferably, the method for determining whether there is an abnormal target individual based on target detection is:

[0033] The monitoring coefficient of the i-th target camera The preset monitoring coefficient threshold interval To compare the sizes:

[0034] like > , then it is determined that there is an abnormal target individual;

[0035] like ≥ ≥ , then it is determined that there is an abnormal individual risk and continuous monitoring is carried out;

[0036] like < , then it is determined that there are no abnormal individuals.

[0037] Preferably, the target frame annotation and the method of obtaining multiple image frames containing the target frame annotation are as follows:

[0038] Let the prediction box be , the real frame is , through the formula Get the prediction box and real frame The target coefficient ;

[0039] in, For the prediction box Upper left corner coordinates; For the prediction box Upper right corner coordinates; For real frame Upper left corner coordinates; For real frame Upper right corner coordinates; For the prediction box and real frame The intersection area value of ; ; ;

[0040] By target coefficient Substitute to obtain different target box annotations and obtain multiple image frames containing target box annotations.

[0041] Preferably, the computer-based intelligent security monitoring platform includes:

[0042] Computer platform of security monitoring terminal, the system transmits the acquired monitoring information to the security monitoring terminal:

[0043] The process of obtaining monitoring information is as follows:

[0044] The information collection unit collects the types of safety events and dangerous events in the historical database for pre-processing; and is used to collect real-time monitoring data obtained by multiple cameras;

[0045] The information processing unit filters and obtains abnormal behavior information according to the pre-processed security event type and dangerous event type and marks it; and is used to identify abnormal behavior information of multiple cameras respectively;

[0046] The network construction unit determines the camera that has identified abnormal behavior information as a target camera and constructs a target camera network;

[0047] The security monitoring unit obtains a set of real-time monitoring data of all target cameras in the target camera network within a period of time;

[0048] The information analysis unit performs target detection analysis and target frame annotation on the real-time monitoring data set obtained by multiple target cameras, and obtains multiple image frames containing target frame annotations, and determines whether there are abnormal target individuals in the target frame based on target detection:

[0049] If so, obtain the characteristic coding sequence of the abnormal target individual under multiple different cameras;

[0050] If not, continue with security monitoring;

[0051] The information fusion unit performs multimodal fusion matching on the feature coding sequence of the abnormal target individual based on the feature matching algorithm, obtains the matched camera target trajectory fragment, constructs the movement trajectory information of the abnormal target individual, and updates it to the security monitoring terminal in real time.

[0052] The beneficial effects of the present invention are as follows: the present invention constructs a target camera network by screening and acquiring abnormal behavior information according to the types of security events and dangerous events, thereby realizing accurate identification of target cameras; the present invention performs target detection analysis and target frame annotation on a real-time monitoring data set acquired by multiple target cameras through an information analysis unit, and obtains multiple image frames containing target frame annotations, thereby realizing accurate positioning of the target frame and synchronous identification of abnormal target individuals; and performs multimodal fusion matching on the feature coding sequences of the abnormal target individuals based on the feature matching algorithm, thereby obtaining matched camera target trajectory fragments, constructing the movement trajectory information of the abnormal target individuals, and realizing the abnormal information identification process of computer intelligent security monitoring.

[0053] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0055] Figure 1 This is a unit diagram of a computer-based intelligent security monitoring system according to the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] The existing computer intelligent security monitoring has the problem of poor synchronization between the camera risk capture status and the camera's monitoring image recognition risk information, and the camera network construction has the problem of insufficient risk capture, incomplete risk range feedback and untimely risk information reception. To solve this problem, please refer to Figure 1 As shown, the present invention is a computer-based intelligent security monitoring system, comprising:

[0058] An information collection unit, used to collect the types of safety events and dangerous events in the historical database for pre-processing; and to collect real-time monitoring data obtained by multiple cameras;

[0059] An information processing unit, used to filter, obtain and mark abnormal behavior information according to the pre-processed security event type and dangerous event type; and to identify abnormal behavior information for multiple cameras respectively;

[0060] A network construction unit, used to determine the camera that has identified abnormal behavior information as a target camera and build a target camera network;

[0061] A security monitoring unit is used to obtain a collection of real-time monitoring data of all target cameras in a target camera network over a period of time;

[0062] The information analysis unit is used to perform target detection analysis and target frame annotation on the real-time monitoring data set obtained by multiple target cameras, and obtain multiple image frames containing target frame annotations, and determine whether there are abnormal target individuals in the target frame according to target detection:

[0063] If so, obtain the characteristic coding sequence of the abnormal target individual under multiple different cameras;

[0064] If not, continue with security monitoring;

[0065] The information fusion unit is used to perform multimodal fusion matching on the feature coding sequence of the abnormal target individual based on the feature matching algorithm, obtain the matched camera target trajectory fragment, construct the movement trajectory information of the abnormal target individual, and update it to the security monitoring terminal in real time.

[0066] In the above technical solution, this embodiment realizes the intelligent analysis process of abnormal behavior analysis and image target detection in the security system by designing a computer-based intelligent security monitoring system, and through the core technology of artificial intelligence: the development of cutting-edge technologies such as machine learning, deep learning, neural networks, etc., and the important role of this intelligent security technology in public places such as shopping malls, airports, and stations, as well as special security projects in prisons, schools, hospitals, etc. This embodiment realizes the optimization and improvement process of monitoring accuracy of common intelligent scenes through the following units.

[0067] First, by setting up the information collection unit to collect the types of security events and dangerous events in the historical database, the dangerous event types in these event classes include abnormal behaviors of people coming and going in the monitoring images of important sites, including security issues of the monitoring sites, security issues including security accidents such as theft and fire, and dangerous abnormal situations such as sudden fainting of personnel and abnormal smoking of equipment. Other daily routine inspections of these dangerous event types are set as security event types; these security event types and dangerous event types are pre-processed to classify and organize the data, remove redundant information, and provide an accurate and concise data basis for subsequent processing; the information collection unit first focuses on. These event types cover. During the preprocessing process, it will perform key features on these data.

[0068] For example, for theft incidents among the dangerous event types, it will sort out key information such as the time, location, and type of stolen items of the theft. For fire incidents among the dangerous event types, it will focus on key information such as the location of the fire source and the direction of fire spread.

[0069] At the same time, the information collection unit is also responsible for obtaining real-time monitoring data from multiple cameras. These cameras are distributed in different monitoring areas, such as various floors of buildings, different areas of parking lots, and various corners of the park. They can simultaneously receive video stream data from multiple cameras. These data contain rich visual information and are an important source for joint discovery of real-time abnormal situations.

[0070] Second, by setting up an information processing unit, based on the pre-processed security event type and dangerous event type, abnormal behavior information is screened from a large amount of monitoring data; for example, if historical data shows that a certain area is prone to theft during a certain period of time, then when a person wanders around or snoops in the area for a long time, it will be marked as abnormal behavior information. For the marking process, specific identification symbols or data tags are generally used to distinguish these abnormal behavior information from normal behavior for subsequent identification and processing.

[0071] The information processing unit also identifies abnormal behavior information of multiple cameras separately to ensure the identification process of the information and location of cameras that often have abnormal behavior information; due to the different environments and monitoring ranges of different cameras, the abnormal behaviors that may occur are also different; through targeted identification of each camera, abnormal situations can be discovered more accurately.

[0072] 3. By setting up a network construction unit, the camera that recognizes abnormal behavior information is determined as the target camera; the target camera is a key node in the entire security monitoring system, and the images it captures contain abnormal target information. For example, when a camera finds a suspicious person climbing over a fence, the camera will be marked as a target camera. Of course, the specific target camera information needs to be determined according to the degree of abnormality in the camera according to the security level. Then, after determining the target camera, the target camera network is constructed. The camera resources related to the abnormal situation are integrated to ensure the joint, comprehensive and timely discovery of abnormal dangers, so as to better track and monitor the development of abnormal events. In addition, in the actual network construction process, factors such as the geographical location relationship between cameras and the overlap of monitoring ranges will be considered, so that these target cameras can work together to provide a more comprehensive monitoring perspective.

[0073] 4. Set up a security monitoring unit to obtain a set of real-time monitoring data of all target cameras in the target camera network within a period of time. Set the time interval according to specific monitoring needs. For example, in a high-security area, a shorter time interval can be set to obtain more timely monitoring information; in a relatively safe area, a longer time interval can be set. The real-time monitoring data set obtained contains the video images captured by multiple target cameras within a period of time, which is the basis for subsequent information identification and analysis.

[0074] 5. By setting up an information analysis unit, we first perform target detection analysis on the real-time monitoring data set obtained by multiple target cameras. Through the visual algorithm, we can identify various targets in the monitoring screen, such as people, vehicles, objects, etc. Then, we mark the identified targets with target frames, and frame the targets with rectangular frames set at the same time and in the same scene to clarify the location and range of the targets. For more complex monitoring scenes, such as crowded public places or parking lots with many vehicles, the algorithm needs to have high-precision recognition capabilities to ensure that each target is accurately marked.

[0075] In addition, the information analysis unit also determines whether there is an abnormal target individual in the target frame based on the target detection result. The specific analysis process is performed by combining the abnormal behavior information marked by the information processing unit and the specific monitoring scene requirements. For example, if a vehicle target frame is found in an area where vehicles are prohibited from passing, the vehicle may be determined to be an abnormal target individual; if a person is found to be engaging in aggressive behavior in a passage where people normally pass, he or she is determined to be an abnormal target individual. If it is determined that there is an abnormal target individual, the feature coding sequence of the abnormal target individual under multiple different cameras is obtained, which contains various coding information of the abnormal target individual such as appearance characteristics, behavioral characteristics, etc., which is an important basis for subsequent feature matching; if it is determined that there is no abnormal target individual, security monitoring continues to maintain continuous and synchronous monitoring of the monitoring screen.

[0076] Sixth, by setting up an information fusion unit, it performs multimodal fusion matching on the feature coding sequence of the abnormal target individual based on the feature matching algorithm. The multimodal fusion matching here refers to comprehensively considering multiple feature information of the abnormal target individual, such as visual features, behavioral pattern features, etc., and fusing and matching these different modal features through complex algorithms. For example, the appearance features of the abnormal target individual obtained from cameras at different angles are matched with the behavioral trajectory features to improve the accuracy of identifying the abnormal target individual.

[0077] Through multimodal fusion matching, the matched camera target trajectory fragments are obtained, and then the movement trajectory information of the abnormal target individual is constructed; the movement path of the abnormal target individual in the monitoring area can be clearly displayed through the movement trajectory information, which is convenient for tracking the whereabouts of the abnormal target individual and predicting its possible behavior. Finally, by updating the constructed movement trajectory information to the security monitoring terminal in real time, the security monitoring personnel can obtain the latest monitoring situation in time and make corresponding decisions.

[0078] As an embodiment of the present invention, the method in which the information processing unit screens and obtains abnormal behavior information and marks it according to the pre-processed security event type and dangerous event type is:

[0079] Obtain samples of security event types and dangerous event types in security monitoring based on big data;

[0080] Extracting three-dimensional data sets of key points from image information of safety event type samples and dangerous event type samples respectively;

[0081] The three-dimensional data set of safety event type samples is collected as positive samples, and the three-dimensional data set of dangerous event type samples is collected as negative samples and input into the convolutional neural network model for learning to obtain a trajectory recognition model;

[0082] The three-dimensional data set corresponding to the real-time monitoring data obtained by multiple cameras is input into the trajectory recognition model to judge abnormal behavior, obtain abnormal behavior information in security monitoring and mark it.

[0083] In the above technical scheme, the specific method for obtaining and marking abnormal behavior information in the information processing unit of this embodiment is to first obtain historical data in security monitoring in a large database, and the historical data includes security event type samples and dangerous event type samples; then extract three-dimensional data sets of key points in the image information of the security event type samples and dangerous event type samples respectively; the three-dimensional data set includes the three-dimensional coordinates of the vertices of the outer contour of the image information, usually by calculating the contour vertex farthest from the geometric center of the outer contour of the image; then, the three-dimensional data set of the security event type samples is taken as a positive sample, and the three-dimensional data set of the dangerous event type samples is taken as a negative sample and input into the convolutional neural network model for learning to obtain a trajectory recognition model; the convolutional neural network model is trained multiple times to optimize the model, and finally the three-dimensional data set corresponding to the real-time monitoring data obtained by multiple cameras is input into the trajectory recognition model for abnormal behavior judgment, and the abnormal behavior information in the security monitoring is obtained and marked; the monitoring data is obtained in time through the trajectory recognition model to realize the accurate capture process of abnormal behavior information.

[0084] As an implementation mode of the present invention, a method of inputting a three-dimensional data set corresponding to the real-time monitoring data acquired by a camera into a trajectory recognition model for abnormal behavior judgment is as follows:

[0085] By formula Calculate the recognition coefficient of the i-th camera ;

[0086] in, is the three-dimensional data function of the i-th camera; is the k-th 3D data function of the i-th camera Axis coordinates; is the k-th 3D data function of the i-th camera Axis coordinates; is the k-th 3D data function of the i-th camera Axis coordinates; is the standard three-dimensional data function of the i-th camera; is the standard 3D data function of the i-th camera Axis coordinates; is the standard 3D data function of the i-th camera Axis coordinates; is the standard 3D data function of the i-th camera Axis coordinates; is the average 3D data function of historical cameras; is the average 3D data function of the historical camera Axis coordinates; is the average 3D data function of the historical camera Axis coordinates; is the average 3D data function of the historical camera Axis coordinates; is the number of three-dimensional data function terms, and .

[0087] In the above technical solution, in this embodiment, the identification coefficient The process of timely capturing the movement trajectory of the camera is realized by obtaining the corresponding three-dimensional data from different cameras, judging the changes of the key points of the image in the current camera by judging the cumulative change state of the three-dimensional data function, and using the size range of the recognition coefficient to judge whether there is a risk of abnormal behavior information in the operation of the camera.

[0088] What needs to be explained is that the three-dimensional data function It is a function set based on the position relationship of different key points in the trajectory recognition model based on the three-dimensional data set input, and is fitted based on the coordinate information of the point position, and changes with the change of the coordinate information of the key point; standard three-dimensional data function , , , All of them are obtained by fitting historical experience data, which will not be described in detail here; and the average parameters , , , All settings are selected based on historical experience data and will not be elaborated here.

[0089] As an implementation mode of the present invention, the abnormal behavior information recognition method for multiple cameras is to use the recognition coefficient of the i-th camera The threshold interval of the standard recognition coefficient To compare:

[0090] like > , then it is judged that the current running trajectory of the i-th camera has the risk of abnormal behavior information, and the camera information is collected;

[0091] like ∈ , then it is judged that the risk of abnormal behavior information of the current running trajectory of the i-th camera is low;

[0092] like < , then it is judged that there is no abnormal behavior information in the current running trajectory of the i-th camera.

[0093] As an embodiment of the present invention, the target detection analysis method of the information analysis unit is:

[0094] By formula Calculate the monitoring coefficient of the i-th target camera ;

[0095] in, is the total number of individual target features detected by the target camera, and ∈ ; is the jth target feature distance parameter of the i-th target camera individual; is the standard distance parameter of the jth target feature of the i-th target camera individual; is the center distance parameter of the jth target feature of the individual; is the preset weight coefficient of the jth target feature of the i-th target camera individual.

[0096] In the above technical solution, in this embodiment, a target detection and analysis method is designed to obtain whether there is an abnormality in the monitoring content of the target camera. Taking the individual person identified in security monitoring as an example, the formula Calculate the monitoring coefficient of the i-th target camera The monitoring coefficient can reflect the changes in the camera's monitoring target. The distance change state of the monitoring target within the target frame in the monitoring screen can show the individual state in detail, and can identify abnormal information, and then timely identify and warn of abnormal individual information.

[0097] in, , All of them are obtained by fitting historical experience data and will not be described in detail here; target feature distance parameters The fitting parameter of the distance between the geometric center of the target frame of the current camera recognition screen and the target feature; The fitting parameter for the standard distance between the geometric center of the target frame of the current camera recognition image and the target feature; the center distance parameter It is to set the geometric center of the individual to be identified in the target frame of the camera recognition screen, and use the distance between the standard target frame in the individual and the geometric center as the center distance parameter.

[0098] As an implementation of the present invention, a method for determining whether there is an abnormal target individual based on target detection is:

[0099] The monitoring coefficient of the i-th target camera The preset monitoring coefficient threshold interval To compare the sizes:

[0100] like > , then it is determined that there is an abnormal target individual;

[0101] like ≥ ≥ , then it is determined that there is an abnormal individual risk and continuous monitoring is carried out;

[0102] like < , then it is determined that there are no abnormal individuals.

[0103] As an implementation manner of the present invention, a method of performing target frame annotation and obtaining multiple image frames containing target frame annotations is as follows:

[0104] Let the prediction box be , the real frame is , through the formula Get the prediction box and real frame The target coefficient ;

[0105] in, For the prediction box Upper left corner coordinates; For the prediction box Upper right corner coordinates; For real frame Upper left corner coordinates; For real frame Upper right corner coordinates; For the prediction box and real frame The intersection area value of ; ; ;

[0106] By target coefficient Substitute to obtain different target box annotations and obtain multiple image frames containing target box annotations.

[0107] In the above technical scheme, the method of performing target frame annotation and obtaining multiple image frames containing target frame annotations in this embodiment is to obtain the target coefficient by performing overlapping union and intersection analysis on the predicted frame and the real frame, collect the center of the target frame, identify its position and mark it, so as to facilitate real-time annotation of the target frame and accurate identification of related monitoring information.

[0108] As an embodiment of the present invention, the computer-based intelligent security monitoring platform includes:

[0109] Computer platform of security monitoring terminal, the system transmits the acquired monitoring information to the security monitoring terminal:

[0110] The process of obtaining monitoring information is as follows:

[0111] The information collection unit collects the types of safety events and dangerous events in the historical database for pre-processing; and is used to collect real-time monitoring data obtained by multiple cameras;

[0112] The information processing unit filters and obtains abnormal behavior information according to the pre-processed security event type and dangerous event type and marks it; and is used to identify abnormal behavior information of multiple cameras respectively;

[0113] The network construction unit determines the camera that has identified abnormal behavior information as a target camera and constructs a target camera network;

[0114] The security monitoring unit obtains a set of real-time monitoring data of all target cameras in the target camera network within a period of time;

[0115] The information analysis unit performs target detection analysis and target frame annotation on the real-time monitoring data set obtained by multiple target cameras, and obtains multiple image frames containing target frame annotations, and determines whether there are abnormal target individuals in the target frame based on target detection:

[0116] If so, obtain the characteristic coding sequence of the abnormal target individual under multiple different cameras;

[0117] If not, continue with security monitoring;

[0118] The information fusion unit performs multimodal fusion matching on the feature coding sequence of the abnormal target individual based on the feature matching algorithm, obtains the matched camera target trajectory fragment, constructs the movement trajectory information of the abnormal target individual, and updates it to the security monitoring terminal in real time.

[0119] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0120] The above contents are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined in this application, they shall all fall within the protection scope of the present invention.

Claims

1. A computer-based intelligent security monitoring system, characterized in that: include: An information collection unit, used to collect the types of safety events and dangerous events in the historical database for pre-processing; and for collecting real-time monitoring data obtained by multiple cameras; An information processing unit, used to filter and obtain abnormal behavior information and mark it according to the pre-processed security event type and dangerous event type; and for identifying abnormal behavior information of multiple cameras respectively; The abnormal behavior information is obtained by inputting the three-dimensional data set corresponding to the real-time monitoring data obtained by the camera into the trajectory recognition model to judge the abnormal behavior, including: By formula Calculate the recognition coefficient of the i-th camera ; in, is the three-dimensional data function of the i-th camera; is the k-th 3D data function of the i-th camera Axis coordinates; is the k-th 3D data function of the i-th camera Axis coordinates; is the k-th 3D data function of the i-th camera Axis coordinates; is the standard three-dimensional data function of the i-th camera; is the standard 3D data function of the i-th camera Axis coordinates; is the standard 3D data function of the i-th camera Axis coordinates; is the standard 3D data function of the i-th camera Axis coordinates; is the average 3D data function of historical cameras; is the average 3D data function of the historical camera Axis coordinates; is the average 3D data function of the historical camera Axis coordinates; is the average 3D data function of the historical camera Axis coordinates; is the number of three-dimensional data function terms, and ; A network construction unit, used to determine the camera that has identified abnormal behavior information as a target camera and build a target camera network; A security monitoring unit is used to obtain a collection of real-time monitoring data of all target cameras in a target camera network over a period of time; The information analysis unit is used to perform target detection analysis and target frame annotation on the real-time monitoring data set obtained by multiple target cameras, and obtain multiple image frames containing target frame annotations, and determine whether there are abnormal target individuals in the target frame according to target detection: If so, obtain the characteristic coding sequence of the abnormal target individual under multiple different cameras; If not, continue with security monitoring; The information fusion unit is used to perform multimodal fusion matching on the feature coding sequence of the abnormal target individual based on the feature matching algorithm, obtain the matched camera target trajectory fragment, construct the movement trajectory information of the abnormal target individual, and update it to the security monitoring terminal in real time.

2. A computer-based intelligent security monitoring system according to claim 1, characterized in that: The method in which the information processing unit screens and obtains abnormal behavior information and marks it according to the pre-processed security event type and dangerous event type is: Obtain samples of security event types and dangerous event types in security monitoring based on big data; Extracting three-dimensional data sets of key points from image information of safety event type samples and dangerous event type samples respectively; The three-dimensional data set of safety event type samples is collected as positive samples, and the three-dimensional data set of dangerous event type samples is collected as negative samples and input into the convolutional neural network model for learning to obtain a trajectory recognition model; The three-dimensional data set corresponding to the real-time monitoring data obtained by multiple cameras is input into the trajectory recognition model to judge abnormal behavior, obtain abnormal behavior information in security monitoring and mark it.

3. According to claim 1, a computer-based intelligent security monitoring system is characterized in that: The method for identifying abnormal behavior information of multiple cameras is to use the recognition coefficient of the i-th camera The threshold interval of the standard recognition coefficient To compare: like > , then it is judged that the current running trajectory of the i-th camera has abnormal behavior information risk, and the camera information is collected; like ∈ , then it is judged that the risk of abnormal behavior information of the current running trajectory of the i-th camera is low; like < , then it is judged that there is no abnormal behavior information in the current running trajectory of the i-th camera.

4. The computer-based intelligent security monitoring system according to claim 1, characterized in that: The target detection analysis method of the information analysis unit is: By formula Calculate the monitoring coefficient of the i-th target camera ; in, is the total number of individual target features detected by the target camera, and ∈ ; is the jth target feature distance parameter of the i-th target camera individual; is the standard distance parameter of the jth target feature of the i-th target camera individual; is the center distance parameter of the jth target feature of the individual; is the preset weight coefficient of the jth target feature of the i-th target camera individual.

5. A computer-based intelligent security monitoring system according to claim 4, characterized in that: The method for judging whether there is an abnormal target individual based on target detection is: The monitoring coefficient of the i-th target camera The preset monitoring coefficient threshold interval To compare the sizes: like > , then it is determined that there is an abnormal target individual; like ≥ ≥ , then it is determined that there is an abnormal individual risk and continuous monitoring is carried out; like < , then it is determined that there are no abnormal individuals.

6. The computer-based intelligent security monitoring system according to claim 1 is characterized in that: The method of performing target frame annotation and obtaining multiple image frames containing target frame annotations is: Let the prediction box be , the real frame is , through the formula Get the prediction box and real frame The target coefficient ; in, For the prediction box Upper left corner coordinates; For the prediction box Upper right corner coordinates; For real frame Upper left corner coordinates; For real frame Upper right corner coordinates; For the prediction box and real frame The intersection area value of ; ; By target coefficient Substitute to obtain different target box annotations and obtain multiple image frames containing target box annotations.

7. A computer-based intelligent security monitoring platform, characterized in that: According to any one of claims 1 to 6, the computer-based intelligent security monitoring system, the platform comprises: Computer platform of security monitoring terminal, the system transmits the acquired monitoring information to the security monitoring terminal: The process of acquiring the monitoring information is as follows: The information collection unit collects the types of safety events and dangerous events in the historical database for pre-processing; and is used to collect real-time monitoring data obtained by multiple cameras; The information processing unit filters and obtains abnormal behavior information according to the pre-processed security event type and dangerous event type and marks it; and is used to identify abnormal behavior information of multiple cameras respectively; The network construction unit determines the camera that has identified abnormal behavior information as a target camera and constructs a target camera network; The security monitoring unit obtains a set of real-time monitoring data of all target cameras in the target camera network within a period of time; The information analysis unit performs target detection analysis and target frame annotation on the real-time monitoring data set obtained by multiple target cameras, and obtains multiple image frames containing target frame annotations, and determines whether there are abnormal target individuals in the target frame based on target detection: If so, obtain the characteristic coding sequence of the abnormal target individual under multiple different cameras; If not, continue with security monitoring; The information fusion unit performs multimodal fusion matching on the feature coding sequence of the abnormal target individual based on the feature matching algorithm, obtains the matched camera target trajectory fragment, constructs the movement trajectory information of the abnormal target individual, and updates it to the security monitoring terminal in real time.

Images