Security early warning system and method based on video content analysis

Abstract

The invention belongs to the technical field of security monitoring. The invention provides a security early warning system and method based on video content analysis. The method comprises the following steps: extracting first behavior data of a target object in a first monitoring video, and performing behavior anomaly analysis on the target object based on the first behavior data to obtain a first behavior anomaly score; determining a plurality of second cameras according to the first behavior abnormal score and the first behavior data, and extracting second behavior data of the target object from a second monitoring video shot by each second camera; and performing anomaly analysis on the first behavior data and the second behavior data by using a first anomaly analysis model to obtain a second behavior anomaly score, and if the second behavior anomaly score is higher than a behavior anomaly threshold, generating and outputting a security early warning signal. According to the invention, whether the target object in the monitoring video has the abnormal behavior is analyzed based on cooperation among multiple cameras, so that the accuracy of abnormal behavior analysis can be remarkably improved, and the false alarm rate is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the field of security monitoring technology, and in particular to a security early warning system and method based on video content analysis. Background Art

[0002] Video-based security warning technology is a technology that uses artificial intelligence algorithms to automatically analyze and process video images. It can identify different people, objects, and environmental conditions, detect abnormal situations in the monitoring screen, and issue alarms and feedback information in real time.

[0003] Existing video-based security warning technology mainly identifies abnormal behavior of the target object in the video shot by a single camera to conduct security warning. However, it is easy to misidentify abnormal behavior when analyzing whether it is abnormal based on the video shot by a single camera. This technical problem needs to be improved. Summary of the invention

[0004] In response to the above technical problems, the present invention provides a security warning method, system, electronic device, computer storage medium and computer program product based on video content analysis to improve the efficiency and usability of security warning based on video content analysis.

[0005] The present invention discloses a security early warning method based on video content analysis, the method comprising the following steps: Extracting first behavior data of a target object in a first surveillance video, and performing a behavior anomaly analysis on the target object based on the first behavior data to obtain a first behavior anomaly score; wherein the first surveillance video is shot by a first camera; Determine a plurality of second cameras according to the first behavior anomaly score and the first behavior data, and extract second behavior data of the target object from second surveillance videos captured by each of the second cameras; The first behavior data and the second behavior data are analyzed for abnormality using a first abnormality analysis model to obtain a second behavior abnormality score. If the second behavior abnormality score is higher than a behavior abnormality threshold, a security warning signal is generated and output.

[0006] Optionally, extracting first behavior data of the target object in the first surveillance video, and performing behavior anomaly analysis on the target object based on the first behavior data to obtain a first behavior anomaly score includes: Inspecting the surveillance videos shot by each camera, determining the surveillance video in which the target object is present as the first surveillance video, and determining the camera that shot the first surveillance video as the first camera; Use a convolutional network to extract features from the first surveillance video to obtain the first behavior data of the target object, use a rule-based second anomaly analysis model to perform an anomaly analysis on the first behavior data to obtain the first behavior anomaly score.

[0007] Optionally, determining a plurality of second cameras according to the first behavior anomaly score and the first behavior data, and extracting second behavior data of the target object from second surveillance videos captured by each of the second cameras includes: Determine the number of collaborations based on the first behavior anomaly score and the first behavior data, and determine the behavior trajectory of the target object based on the first behavior data; A plurality of second cameras corresponding to the collaborative quantity are obtained by screening based on the layout position of the first camera and the behavior trajectory within the surrounding range of the first camera; The third behavior data are respectively extracted from the second monitoring videos captured by each of the second cameras, and each of the third behavior data is matrix-processed and integrated to obtain the second behavior data of the target object.

[0008] Optionally, determining the collaboration quantity according to the first behavior anomaly score and the first behavior data includes: Determine a first collaboration quantity according to the first behavior anomaly score; wherein the greater the first behavior anomaly score, the smaller the first collaboration quantity, and conversely, the smaller the first behavior anomaly score, the greater the first collaboration quantity; parsing the first behavior data to obtain clothing feature data and body movement data of the target object, performing semantic understanding on the body movement data to obtain a number of body behavior interpretations, evaluating the matching degree between each of the body behavior interpretations and the clothing feature data, and determining an adjustment coefficient based on the matching degree; The first cooperative quantity is adjusted to a second cooperative quantity using the adjustment coefficient, and the second cooperative quantity is determined as the cooperative quantity.

[0009] Optionally, before performing semantic understanding on the body motion data, the method further includes: Determining whether the clothing feature data of the target object parsed from the first behavior data belongs to preset clothing feature data; wherein the preset clothing feature data is determined based on the appearance features of the uniforms of security personnel in the security warning area; If the clothing feature data belongs to the preset clothing feature data, a trigger signal for semantically understanding the body motion data is generated; otherwise, a trigger signal for semantically understanding the body motion data is not generated.

[0010] Optionally, the first anomaly analysis model is constructed based on a convolutional network and a Transformer, including a convolutional network submodule and a Transformer submodule, the Transformer submodule includes a first encoder, a second encoder and a decoder, and the first encoder and the second encoder have different structures; Among them, the convolutional network sub-model extracts the comprehensive behavior characteristics of the target object from the first behavior data and the second behavior data; the comprehensive behavior characteristics are respectively input into the first encoder and the second encoder to obtain the first encoder output and the second encoder output respectively; the first encoder output and the second encoder output are respectively input into the decoder to obtain the third behavior anomaly score and the fourth behavior anomaly score respectively, and the third behavior anomaly score and the fourth behavior anomaly score are weightedly calculated to obtain the second behavior anomaly score.

[0011] The present invention also discloses a security warning system based on video content analysis, the system comprising a processing device and a storage device, the computer code stored in the storage device is called and executed by the processing device to implement the following steps: Extracting first behavior data of a target object in a first surveillance video, and performing a behavior anomaly analysis on the target object based on the first behavior data to obtain a first behavior anomaly score; wherein the first surveillance video is shot by a first camera; Determine a plurality of second cameras according to the first behavior anomaly score and the first behavior data, and extract second behavior data of the target object from second surveillance videos captured by each of the second cameras; The first behavior data and the second behavior data are analyzed for abnormality using a first abnormality analysis model to obtain a second behavior abnormality score. If the second behavior abnormality score is higher than a behavior abnormality threshold, a security warning signal is generated and output.

[0012] The present invention also discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement any of the above methods.

[0013] The present invention also discloses a computer storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement any of the above methods.

[0014] The present invention also discloses a computer program product, which includes computer codes. When the computer codes are executed by a processor of an electronic device, any of the above methods is implemented.

[0015] The solution of the present invention can analyze whether a target object in a surveillance video has abnormal behavior based on collaboration between multiple cameras, which can significantly improve the accuracy of abnormal behavior analysis and thereby effectively reduce the false alarm rate in security warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a flow chart of a security early warning method based on video content analysis disclosed in an embodiment of the present invention; Figure 2 is a structural schematic diagram of a first abnormality analysis model disclosed in an embodiment of the present invention; Figure 3 It is a structural schematic diagram of a security warning system based on video content analysis disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following is a description of the implementation of the present application by specific specific embodiments. People familiar with the technology can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0019] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0020] Existing video-based security warning technology mainly identifies abnormal behavior of the target object in the video shot by a single camera to conduct security warning. However, it is easy to misidentify abnormal behavior when analyzing whether it is abnormal based on the video shot by a single camera. This technical problem needs to be improved.

[0021] In response to the above technical issues, such as Figure 1 As shown, the embodiment of the present invention discloses a security warning method based on video content analysis, and the method comprises the following steps: Extracting first behavior data of a target object in a first surveillance video, and performing a behavior anomaly analysis on the target object based on the first behavior data to obtain a first behavior anomaly score; wherein the first surveillance video is shot by a first camera; Determine a plurality of second cameras according to the first behavior anomaly score and the first behavior data, and extract second behavior data of the target object from second surveillance videos captured by each of the second cameras; The first behavior data and the second behavior data are analyzed for abnormality using a first abnormality analysis model to obtain a second behavior abnormality score. If the second behavior abnormality score is higher than a behavior abnormality threshold, a security warning signal is generated and output.

[0022] Compared with the method in the prior art of judging whether the target object has abnormal behavior based only on a surveillance video shot by a single camera, the present invention first extracts the first behavior data of the target object in a first surveillance video shot by a first camera, performs a preliminary abnormal behavior analysis on the target object based on the first behavior data, and obtains the first behavior abnormality score. Then, according to the first behavior abnormality score and the first behavior data, it is determined that a number of second cameras need to be coordinated, and the second behavior data of the target object is extracted from the second surveillance video shot by each second camera. Finally, the first abnormal analysis model is used to perform abnormal analysis on the first behavior data and the second behavior data to obtain the second behavior abnormality score. If the second behavior abnormality score is higher than the behavior abnormality threshold, a security warning signal is generated and output; if the second behavior abnormality score is lower than the behavior abnormality threshold, the target object is determined to be a non-intrusive object, and a security warning signal is not generated and output at this time.

[0023] The above solution of the present invention can be implemented by a local processor or server in the security warning area. The server can be a conventional server, a cloud server, or a distributed server cluster based on cloud computing, which is not specifically limited in the present invention.

[0024] Optionally, extracting first behavior data of the target object in the first surveillance video, and performing behavior anomaly analysis on the target object based on the first behavior data to obtain a first behavior anomaly score includes: Inspecting the surveillance videos shot by each camera, determining the surveillance video in which the target object is present as the first surveillance video, and determining the camera that shot the first surveillance video as the first camera; Use a convolutional network to extract features from the first surveillance video to obtain the first behavior data of the target object, use a rule-based second anomaly analysis model to perform an anomaly analysis on the first behavior data to obtain the first behavior anomaly score.

[0025] In this embodiment, the present invention constructs a first anomaly analysis model and a second anomaly analysis model. The first anomaly analysis model is used to perform a deeper analysis on whether there is an abnormality in the target object based on the first behavior data and the second behavior data, while the second anomaly analysis model is a conventional model constructed based on rule algorithms such as support vector machines, PCA principal component analysis, isolation forests, random forests, etc., which is used to perform a preliminary shallow analysis of the abnormal behavior of the target object in the surveillance video to quickly determine potential targets with abnormal conditions.

[0026] The security warning system conducts inspections on the surveillance videos shot by each camera. The inspection can be periodic or real-time. If a target object enters a certain surveillance video shot by a camera, the surveillance video is determined as the first surveillance video, and the corresponding shooting camera is determined as the first camera. Among them, the target object mainly refers to people, but can also include animals, vehicles, etc.

[0027] Optionally, determining a plurality of second cameras according to the first behavior anomaly score and the first behavior data, and extracting second behavior data of the target object from second surveillance videos captured by each of the second cameras includes: Determine the number of collaborations based on the first behavior anomaly score and the first behavior data, and determine the behavior trajectory of the target object based on the first behavior data; A plurality of second cameras corresponding to the cooperative number are obtained by screening based on the layout position of the first camera and the behavior trajectory within the surrounding range of the first camera; The third behavior data are respectively extracted from the second monitoring videos captured by each of the second cameras, and each of the third behavior data is matrix-processed and integrated to obtain the second behavior data of the target object.

[0028] In this embodiment, when the first behavior anomaly score obtained by analyzing the second anomaly analysis model used for routine inspections is higher than the anomaly threshold, it is preliminarily determined that the target object is abnormal. At this time, surveillance videos taken by multiple second cameras around the first camera are required to assist in verifying and confirming whether the target object is actually abnormal based on the second surveillance videos taken by the second cameras.

[0029] Specifically, the number of collaborations is determined based on the first behavior anomaly score and the first behavior data of the target object, and the specific determination method is explained in detail in the following embodiments; and, based on the first behavior data, the motion characteristics of the target object in the first surveillance video, such as motion speed, motion direction, etc., can be determined, and then the behavior trajectory of the target object can be fitted, and the behavior trajectory includes both the historical behavior trajectory of the target object and the predicted behavior trajectory of the target object in the future period. So far, within the surrounding range of the first camera, based on the layout position of the first camera, several second cameras that can cover the behavior trajectory are screened. Then, in the same way as the aforementioned extraction of the first behavior data, the third behavior data of the target object are respectively extracted from the second surveillance videos shot by these second cameras, and each third behavior data is matrixed, and then the processed behavior feature matrix is ​​integrated into one piece of data, that is, the second behavior data of the target object is obtained.

[0030] The first behavior data and the second behavior data of the target object include the target object's movement speed, movement direction, clothing features, and the target object's physical behavior corresponding to the preset abnormal behavior, and the abnormal behavior includes but is not limited to climbing, poking, squatting and digging, cutting fences, etc. The abnormal behaviors listed above are only for illustration and are not intended to limit the solution of the present invention to only the above-mentioned types of abnormal behaviors. Other suspicious behaviors of the intruder object belong to the abnormal behaviors of the present invention.

[0031] Optionally, determining the collaboration quantity according to the first behavior anomaly score and the first behavior data includes: Determine a first collaboration quantity according to the first behavior anomaly score; wherein the greater the first behavior anomaly score, the smaller the first collaboration quantity, and conversely, the smaller the first behavior anomaly score, the greater the first collaboration quantity; parsing the first behavior data to obtain clothing feature data and body movement data of the target object, performing semantic understanding on the body movement data to obtain a number of body behavior interpretations, evaluating the matching degree between each of the body behavior interpretations and the clothing feature data, and determining an adjustment coefficient based on the matching degree; The first cooperative quantity is adjusted to a second cooperative quantity using the adjustment coefficient, and the second cooperative quantity is determined as the cooperative quantity.

[0032] In this embodiment, the determination of the number of collaborations is divided into the following two parts: First, the first collaborative quantity is determined based on the first behavior anomaly score and the corresponding negative correlation. Specifically, if the first behavior anomaly score of the target object is larger, the more significant the abnormal behavior of the target object is. In this case, there is no need to dispatch more collaborative cameras, that is, the accuracy of anomaly identification can be ensured by shooting video data of the target object with fewer collaborative cameras; and if the first behavior anomaly score of the target object is smaller, the less significant the abnormal behavior of the target object is. In this case, more collaborative cameras need to be dispatched, and these cameras will shoot surveillance videos of the target object in a larger activity space and time from more shooting angles, so as to perform more accurate behavioral anomaly analysis based on the extracted more comprehensive second behavior data.

[0033] At the same time, security personnel are usually deployed in the security warning area, and the uniforms of security personnel are obviously different from those of intruders, so the degree of abnormal behavior of the target object can be assisted in analysis. However, in practice, it is found that intruders will also disguise themselves by wearing uniforms of the same or similar styles as security personnel. Therefore, it is necessary to distinguish the normal patrol behavior of security personnel from the intrusion behavior of intruders, and to assist in determining a more appropriate number of coordinated second cameras. Specifically, the clothing feature data and body movement data of the target object are first parsed from the first behavior data, and the semantic analysis component is used to semantically understand the body movement data to obtain several body behavior explanations, such as using lighting fixtures to look around, pulling fences by hand (for example, security personnel are confirming the strength of the fence), etc. The matching degree of each body behavior interpretation and clothing feature data is evaluated. If the matching degree is high, it means that the target object is more likely to be a security personnel. In this case, the adjustment coefficient is set to a smaller value, and the first coordination quantity is appropriately lowered using the smaller adjustment coefficient. If the matching degree is low, it means that the target object is more likely to be an intruder. In this case, the adjustment coefficient is set to a larger value, and the first coordination quantity is appropriately increased using the larger adjustment coefficient. The corresponding relationship between the adjustment coefficient and the matching degree can be determined by a comparison table or by a conversion formula obtained by fitting, and the present invention does not make specific limitations.

[0034] Therefore, by analyzing the matching degree between the target object's body behavior interpretation and clothing feature data, the appropriate adjustment coefficient is determined, thereby determining the appropriate number of second cameras for collaboration, achieving a certain balance between anomaly analysis accuracy and anomaly analysis efficiency.

[0035] Optionally, before performing semantic understanding on the body motion data, the method further includes: Determining whether the clothing feature data of the target object parsed from the first behavior data belongs to preset clothing feature data; wherein the preset clothing feature data is determined based on the appearance features of the uniforms of security personnel in the security warning area; If the clothing feature data belongs to the preset clothing feature data, a trigger signal for semantically understanding the body motion data is generated; otherwise, a trigger signal for semantically understanding the body motion data is not generated.

[0036] In this embodiment, after the clothing feature data of the target object is parsed from the first behavior data, it is necessary to determine whether it belongs to the preset clothing feature data, that is, to analyze whether it is wearing the same or similar uniform as the security personnel in the security warning area. If the clothing feature data belongs to the preset clothing feature data, it is preliminarily determined that there is a large similarity between it and the security personnel, and a trigger signal for semantic understanding of the body movement data is generated at this time, that is, the adjustment coefficient is used to adjust the first coordination quantity; otherwise, it is determined that the similarity between the target object and the security personnel in clothing is insufficient, and no trigger signal for semantic understanding of the body movement data is generated at this time, that is, the adjustment coefficient is not used to adjust the first coordination quantity.

[0037] Alternatively, if Figure 2 As shown, the first anomaly analysis model is constructed based on a convolutional network and a Transformer, including a convolutional network submodule and a Transformer submodule, the Transformer submodule includes a first encoder, a second encoder and a decoder, and the structures of the first encoder and the second encoder are different; Among them, the convolutional network sub-model extracts the comprehensive behavior characteristics of the target object from the first behavior data and the second behavior data; the comprehensive behavior characteristics are respectively input into the first encoder and the second encoder to obtain the first encoder output and the second encoder output respectively; the first encoder output and the second encoder output are respectively input into the decoder to obtain the third behavior anomaly score and the fourth behavior anomaly score respectively, and the third behavior anomaly score and the fourth behavior anomaly score are weightedly calculated to obtain the second behavior anomaly score.

[0038] In this embodiment, the first anomaly analysis model of the present invention includes a convolutional network submodule and a Transformer submodule, wherein the Transformer submodule adopts a dual encoder architecture, which can significantly improve the confidence of the second behavior anomaly score. Among them, the weighted weights of the third behavior anomaly score and the fourth behavior anomaly score can be preset.

[0039] The first encoder and the second encoder have different structures, which are mainly reflected in the differences in residual connections, the number of encoder layer stacking times, etc.

[0040] like Figure 3 As shown, an embodiment of the present invention further discloses a security warning system based on video content analysis, the system comprising a processing device and a storage device, the computer code stored in the storage device is called and executed by the processing device to implement the following steps: Extracting first behavior data of a target object in a first surveillance video, and performing a behavior anomaly analysis on the target object based on the first behavior data to obtain a first behavior anomaly score; wherein the first surveillance video is shot by a first camera; Determine a plurality of second cameras according to the first behavior anomaly score and the first behavior data, and extract second behavior data of the target object from second surveillance videos captured by each of the second cameras; The first behavior data and the second behavior data are analyzed for abnormality using a first abnormality analysis model to obtain a second behavior abnormality score. If the second behavior abnormality score is higher than a behavior abnormality threshold, a security warning signal is generated and output.

[0041] An embodiment of the present invention further discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method described in the above embodiment.

[0042] An embodiment of the present invention further discloses a computer storage medium, wherein the computer storage medium stores a computer program, and the computer program is executed by a processor to implement the method described in the above embodiment.

[0043] An embodiment of the present invention further discloses a computer program product, which includes computer code. When the computer code is executed by a processor of an electronic device, the method described in the above embodiment is implemented.

[0044] The computer-readable storage medium described above may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. Alternatively, the computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0045] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0046] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0047] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A security early warning method based on video content analysis, characterized in that: The method comprises the following steps: Extracting first behavior data of a target object in a first surveillance video, and performing a behavior anomaly analysis on the target object based on the first behavior data to obtain a first behavior anomaly score; wherein the first surveillance video is shot by a first camera; Determine a plurality of second cameras according to the first behavior anomaly score and the first behavior data, and extract second behavior data of the target object from second surveillance videos captured by each of the second cameras; The first behavior data and the second behavior data are analyzed for abnormality using a first abnormality analysis model to obtain a second behavior abnormality score. If the second behavior abnormality score is higher than a behavior abnormality threshold, a security warning signal is generated and output.

2. The security early warning method based on video content analysis according to claim 1, characterized in that: Extracting first behavior data of a target object in a first surveillance video, and performing a behavior anomaly analysis on the target object based on the first behavior data to obtain a first behavior anomaly score, including: Inspecting the surveillance videos shot by each camera, determining the surveillance video in which the target object is present as the first surveillance video, and determining the camera that shot the first surveillance video as the first camera; Use a convolutional network to extract features from the first surveillance video to obtain the first behavior data of the target object, use a rule-based second anomaly analysis model to perform an anomaly analysis on the first behavior data to obtain the first behavior anomaly score.

3. A security early warning method based on video content analysis according to claim 2, characterized in that: Determining a plurality of second cameras according to the first behavior anomaly score and the first behavior data, and extracting second behavior data of the target object from second surveillance videos captured by each of the second cameras, including: Determine the number of collaborations based on the first behavior anomaly score and the first behavior data, and determine the behavior trajectory of the target object based on the first behavior data; A plurality of second cameras corresponding to the cooperative number are obtained by screening based on the layout position of the first camera and the behavior trajectory within the surrounding range of the first camera; The third behavior data are respectively extracted from the second monitoring videos captured by each of the second cameras, and each of the third behavior data is matrix-processed and integrated to obtain the second behavior data of the target object.

4. The security early warning method based on video content analysis according to claim 3 is characterized in that: Determining the number of collaborations according to the first behavior anomaly score and the first behavior data includes: Determine a first collaboration quantity according to the first behavior anomaly score; wherein the greater the first behavior anomaly score, the smaller the first collaboration quantity, and conversely, the smaller the first behavior anomaly score, the greater the first collaboration quantity; parsing the first behavior data to obtain clothing feature data and body movement data of the target object, performing semantic understanding on the body movement data to obtain a number of body behavior interpretations, evaluating the matching degree between each of the body behavior interpretations and the clothing feature data, and determining an adjustment coefficient based on the matching degree; The first cooperative quantity is adjusted to a second cooperative quantity using the adjustment coefficient, and the second cooperative quantity is determined as the cooperative quantity.

5. The security early warning method based on video content analysis according to claim 4 is characterized in that: Before performing semantic understanding on the body motion data, the method further includes: Determining whether the clothing feature data of the target object parsed from the first behavior data belongs to preset clothing feature data; wherein the preset clothing feature data is determined based on the appearance features of the uniforms of security personnel in the security warning area; If the clothing feature data belongs to the preset clothing feature data, a trigger signal for semantically understanding the body motion data is generated; otherwise, a trigger signal for semantically understanding the body motion data is not generated.

6. A security early warning method based on video content analysis according to claim 5, characterized in that: The first anomaly analysis model is constructed based on a convolutional network and a Transformer, and includes a convolutional network submodule and a Transformer submodule, wherein the Transformer submodule includes a first encoder, a second encoder and a decoder, and the structures of the first encoder and the second encoder are different; Among them, the convolutional network sub-model extracts the comprehensive behavior characteristics of the target object from the first behavior data and the second behavior data; the comprehensive behavior characteristics are respectively input into the first encoder and the second encoder to obtain the first encoder output and the second encoder output respectively; the first encoder output and the second encoder output are respectively input into the decoder to obtain the third behavior anomaly score and the fourth behavior anomaly score respectively, and the third behavior anomaly score and the fourth behavior anomaly score are weightedly calculated to obtain the second behavior anomaly score.

7. A security warning system based on video content analysis, the system comprising a processing device and a storage device, characterized in that: The computer code stored in the storage device is called and executed by the processing device to implement the following steps: Extracting first behavior data of a target object in a first surveillance video, and performing a behavior anomaly analysis on the target object based on the first behavior data to obtain a first behavior anomaly score; wherein the first surveillance video is shot by a first camera; Determine a plurality of second cameras according to the first behavior anomaly score and the first behavior data, and extract second behavior data of the target object from second surveillance videos captured by each of the second cameras; The first behavior data and the second behavior data are analyzed for abnormality using a first abnormality analysis model to obtain a second behavior abnormality score. If the second behavior abnormality score is higher than a behavior abnormality threshold, a security warning signal is generated and output.

8. An electronic device, comprising: At least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method according to any one of claims 1 to 6.

9. A computer storage medium storing a computer program, characterized in that: The computer program is executed by a processor to implement the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that: The computer program product includes computer codes, and when the computer codes are executed by a processor of an electronic device, the method according to any one of claims 1 to 6 is implemented.

Images