Active alarm information generation method and system based on AI video analysis

Abstract

The invention discloses an active alarm information generation method and system based on AI video analysis, and belongs to the technical field of video surveillance, and the method specifically comprises the steps: creating a smoke and fire detection task, selecting a corresponding video analysis algorithm, and drawing a recognition region; adjusting the depth of field of a camera lens to be shallow depth of field, blurring the recognition area, and marking blurred light spots of a known light source; when an unmarked blurred light spot appears in the identification area, identifying the color temperature value of the blurred light spot and the maximum diameter of the light spot, marking the blurred light spot with the color temperature value and the light spot conforming to the setting as an undetermined light spot, and marking the minimum external rectangular area of the undetermined light spot as a target area; adjusting the depth of field of the camera lens to be deep depth of field, collecting the image frame again, cutting the image block corresponding to the target area, inputting the image block into the preset image recognition model, outputting the recognition result of the image block, triggering alarm pushing when the recognition result is smoke and fire, and otherwise, continuing monitoring.

Description

Technical Field

[0001] The present invention relates to the field of video surveillance technology, and in particular to a method and system for generating active alarm information based on AI video analysis. Background Art

[0002] With the rapid development of artificial intelligence technology, especially the continuous improvement of deep learning methods and performance, technologies such as computer vision, image processing, video structuring and data analysis are also constantly improving. The advancement of these technologies has provided a solid technical foundation for the application of AI video analysis. In practical applications, traditional video surveillance systems often face the problems of "difficult extraction and difficult search", unreliable target feature retrieval, large image search and comparison errors, and a lack of effective methods for deep mining of video image information. These problems limit the application value of video surveillance systems, and AI video analysis can solve these problems and improve the efficiency and accuracy of video surveillance.

[0003] During AI video analysis, the system needs to extract useful feature information from the entire photo to facilitate subsequent classification or recognition tasks. However, the number of features contained in the entire photo is large and complex, which will lead to a large demand for image recognition computing power, and the entire photo may contain a large amount of irrelevant information, which will interfere with the recognition results and reduce the accuracy of recognition. At the same time, image recognition of the entire photo will also increase the complexity and time cost of the calculation. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for generating active warning information based on AI video analysis to solve the following technical problems:

[0005] The number of features contained in the entire photo is huge and complex, which will result in a large demand for computing power for image recognition. In addition, the entire photo may contain a large amount of irrelevant information, which will interfere with the recognition results and thus reduce the accuracy of recognition.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A method for generating active warning information based on AI video analysis, comprising the following steps:

[0008] Select a camera from the camera list as the monitoring object, create a fire and smoke detection task, set the task execution time, select the corresponding video analysis algorithm in the algorithm list and draw the ROI recognition area; set the corresponding video frame rate;

[0009] Adjust the depth of field of the camera lens to a shallow depth of field to blur the ROI recognition area, mark the blurred light spots of known light sources in the ROI recognition area, and perform real-time light spot detection on the ROI recognition area according to the video frame extraction frequency;

[0010] When an unmarked blurred light spot appears in the ROI identification area, the color temperature value and the maximum diameter of the blurred light spot are identified, and the blurred light spot with a color temperature value within the set range and a maximum diameter less than the set threshold is marked as a pending light spot, and the minimum circumscribed rectangular area of ​​the pending light spot is marked as the target area;

[0011] Adjust the camera lens depth of field to deep depth of field, collect the ROI recognition area image frame again, crop the image block corresponding to the target area, input the image block into a preset image recognition model, output the recognition result of the image block, and trigger an alarm push when the recognition result is fireworks, otherwise continue to perform spot detection on the ROI recognition area.

[0012] As a further solution of the present invention, the process of adjusting the depth of field of the camera lens to a shallow depth of field is as follows:

[0013] The distance between the focal plane of the camera and the camera lens is adjusted so that the distance is within the interval [A, A+a], where A is the minimum focusing distance of the camera and a is the set redundancy value.

[0014] As a further solution of the present invention: the process of adjusting the depth of field of the camera lens to a deep depth of field is:

[0015] Adjust the aperture value of the camera lens so that the entire ROI recognition area is within the imaging depth of field of the camera lens.

[0016] As a further solution of the present invention: the process of real-time spot detection of the ROI identification area is:

[0017] A selection box is set according to the shape of the lens aperture blades, and the deformation range of the selection box is set. The image frame in the ROI recognition area is edge detected and binarized to obtain a binary image. The marked blurred light spot is excluded from the binary image. The contour of the connected area with a gray value of 255 in the binary area is selected based on the selection box, and the contour within the deformation range of the selection box is screened out and marked as the light spot contour. The light spot contour and the internal area are the blurred light spot.

[0018] As a further solution of the present invention: the process of identifying the color temperature value of the blurred light spot is:

[0019] For any blurred light spot, when the color of the blurred light spot is white, the color temperature of the blurred light spot is the white balance parameter of the current camera;

[0020] When the color of the blurred light spot is light blue or blue, gradually increase the white balance parameter of the camera until the color of the blurred light spot turns white. The color temperature of the blurred light spot is the white balance parameter of the current camera.

[0021] When the color of the blurred light spot is yellow or orange, the white balance parameter of the camera is gradually reduced until the color of the blurred light spot becomes white, and the color temperature of the blurred light spot is the white balance parameter of the current camera.

[0022] As a further solution of the present invention: the process of identifying the maximum diameter of the blurred light spot is:

[0023] For any blurred light spot, select several points in the outline of the blurred light spot at equal intervals, take each point as the starting point, draw a connecting line between the starting point and the other side of the outline through the centroid of the outline, and select the longest connecting line to mark it as the maximum diameter of the blurred light spot.

[0024] As a further solution of the present invention, the specific process of marking the blurred light spot whose color temperature value is within the set range and whose maximum diameter is less than the set threshold as the pending light spot is:

[0025] Obtain the flame color temperature K of the fireworks to be detected, set the error threshold k, and then the setting range is [Kk, K+k];

[0026] The fireworks to be detected are set at the farthest distance in the identification area, the camera lens is kept at a shallow depth of field, and the blurred spot size of the fireworks in the camera image is recorded at this time, and this size is used as the set threshold.

[0027] The present invention also includes a proactive warning information generation system based on AI video analysis, which is used to implement the proactive warning information generation method based on AI video analysis, including:

[0028] The task configuration module is used to create a fire and smoke detection task, set the task execution time, select the corresponding video analysis algorithm in the algorithm list and draw the ROI recognition area; set the corresponding video frame extraction frequency;

[0029] A depth of field adjustment module is used to adjust the depth of field of the camera lens to a shallow depth of field or a deep depth of field;

[0030] The light spot monitoring module is used to mark the blurred light spots of known light sources in the ROI identification area, and perform real-time light spot detection on the ROI identification area according to the video frame extraction frequency; when an unmarked blurred light spot appears in the ROI identification area, the color temperature value and the maximum diameter of the blurred light spot are identified, and the blurred light spot with a color temperature value within a set range and a maximum diameter less than a set threshold is marked as a pending light spot, and the minimum circumscribed rectangular area of ​​the pending light spot is marked as the target area;

[0031] The image recognition module is used to collect the image frame of the ROI recognition area, cut the image block corresponding to the target area, input the image block into a preset image recognition model, and output the recognition result of the image block.

[0032] The alarm prompt module is used to trigger an alarm push when the recognition result is fireworks, otherwise continue to perform spot detection on the ROI recognition area.

[0033] Beneficial effects of the present invention:

[0034] The present invention significantly reduces the amount of data to be processed by first identifying the blurred light spots of fireworks and cutting out the target area, and then performing image recognition, thereby reducing the computing power requirements of the system. This method avoids complex image recognition processing of the entire photo, so that computing resources are more effectively utilized. Through the precise positioning of the target area and the identification of the blurred light spots, the interference of irrelevant information can be effectively eliminated, thereby improving the accuracy of subsequent image recognition. In particular, when detecting specific targets such as fireworks, this targeted processing method can significantly improve the reliability of the recognition results. After cutting out the target area, only this smaller area is image recognized, which greatly reduces the complexity and time cost of the calculation. This not only improves the processing efficiency of the system, but also enables the system to complete the recognition task in a shorter time, meeting the needs of real-time monitoring. Due to the reduction in computational complexity and the improvement in processing speed, the method of the present invention can generate warning information more quickly, enhancing the real-time response capability of the system. This is crucial for timely discovering potential safety hazards and taking preventive measures in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further described below in conjunction with the accompanying drawings.

[0036] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0037] 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.

[0038] See also Figure 1 As shown, the present invention is a method for generating active warning information based on AI video analysis, comprising the following steps:

[0039] 1. Select one or more cameras from the camera list as the objects for monitoring and analysis. Create a task and fill in the task name: Fireworks detection task, that is, smoking detection. Click the "Add Task" button to enter the task creation page. Next, select a suitable video analysis service node according to your needs. This node will be responsible for processing and analyzing the video data from the camera. Set the execution time and frequency plan parameters of the task according to actual needs and select the task plan. In the algorithm list, select the corresponding video analysis algorithm and draw the ROI recognition area. Set a reasonable video frame extraction frequency according to analysis requirements. This helps to reduce resource consumption while ensuring analysis accuracy. After completing all the above settings, click the "Save" button to successfully create the task.

[0040] Task start and close: You can quickly find the tasks you want to manage by using the query and filter functions of the task list. In the task list, select one or more tasks you want to start or close. Click the "Start" or "Close" button to perform the corresponding operation on the selected task.

[0041] Batch modify configuration: For easy management, the Task Scheduling Center also provides the function of batch modify configuration. Similarly, through the query and filter function of the task list, find the task you want to modify the configuration. Select one or more tasks that need to modify the configuration. Click the "Batch modify configuration" button to enter the configuration modification page. On this page, you can batch modify the default frame rate, default recognition delay and other parameters of the selected tasks.

[0042] Batch configuration plan: In addition, the Task Scheduling Center also supports the function of batch configuration plan. Use the query and filter function of the task list to find the task for which you want to set the analysis time schedule. Select one or more tasks for which you want to set the analysis time schedule. Click the "Batch Configuration Plan" button to enter the plan setting page. On this page, you can batch set the analysis time schedule for the selected tasks to ensure that they can perform the video algorithm analysis work according to the scheduled time.

[0043] 2. Adjust the depth of field of the camera lens to a shallow depth of field. Adjusting the depth of field of the camera lens to a shallow depth of field is a key step. Specifically, this involves adjusting the distance between the focal plane of the camera and the camera lens so that this distance is within a specific interval [A, A+a]. Among them, A represents the closest focusing distance of the camera, and a is a pre-set redundant value to ensure the accuracy and flexibility of the adjustment. Through such adjustments, the image in the ROI (region of interest) identification area will be slightly blurred, which will help the subsequent spot detection process.

[0044] In the ROI recognition area, the system will first mark the blurred light spots produced by known light sources. These spots may be caused by lights, reflections or other light sources in the monitoring environment. The purpose of marking these blurred light spots is to exclude them in the subsequent light spot detection process to avoid them from interfering with the detection results.

[0045] Next, the system will perform real-time spot detection on the ROI recognition area according to the video frame extraction frequency. The video frame extraction frequency refers to the frequency of extracting frames from the video stream for detection, which determines the real-time and accuracy of the system. By setting a suitable frame extraction frequency, the system can ensure that important information is not missed and performance will not be degraded due to processing too much data.

[0046] When performing spot detection, the system will set a selection box based on the shape of the lens aperture blades and set the deformation range of this selection box. This selection box is used to perform edge detection on the image frame of the ROI recognition area and obtain a binary image through binarization processing. In this process, the system will exclude the marked blurred spots from the binary image to prevent them from affecting subsequent processing.

[0047] Then, the system selects the contours of the connected areas with a grayscale value of 255 in the binary area. These connected areas usually correspond to bright parts in the image, such as light spots or reflective points. The system selects the contours within the deformation range of the selection box and marks them as light spot contours. Finally, the light spot contour and the area inside it are determined as blurred light spots.

[0048] 3. When unmarked blurred light spots appear in the ROI identification area, a series of steps are performed to identify the color temperature value and maximum diameter of these blurred light spots.

[0049] For the color temperature value recognition process of the blurred light spot, the system will make a judgment based on the color of the blurred light spot. If the color of the blurred light spot is white, then the color temperature of the blurred light spot is the white balance parameter of the current camera. This is because white usually corresponds to a neutral color temperature, that is, there is no obvious color cast.

[0050] If the color of the blurred light spot is light blue or blue, the system will gradually increase the white balance parameter of the camera until the color of the blurred light spot becomes white. In this process, the system will record the white balance parameter that makes the blurred light spot color become white and use it as the color temperature value of the blurred light spot.

[0051] On the contrary, if the color of the blurred light spot is yellow or orange, the system will gradually reduce the white balance parameter of the camera until the color of the blurred light spot becomes white. Similarly, the system will record the white balance parameter that makes the blurred light spot color become white and use it as the color temperature value of the blurred light spot.

[0052] The next step is to identify the maximum diameter of the blurred light spot. For any blurred light spot, the system will select a number of points at equal intervals on its contour line. Then, starting from each point, a connecting line is drawn between the starting point and the other side of the contour through the contour centroid. Among these connecting lines, the system will select the longest one as the maximum diameter of the blurred light spot.

[0053] For the identified blurred light spots, the system will further process them to determine whether they are potential fireworks targets. Specifically, the system will mark the blurred light spots with color temperature values ​​within the set range and the maximum diameter of the light spots less than the set threshold as pending light spots.

[0054] In order to set this range and threshold, the system first needs to obtain the flame color temperature K of the fireworks to be detected. This color temperature value is usually predetermined based on the type of fireworks and the burning conditions. Then, the system sets an error threshold k to define the acceptable range of color temperature values. In this way, the set range is determined to be [Kk, K+k]. Only when the color temperature value of the blurred light spot falls within this range will it be considered as a possible fireworks target.

[0055] Next, the system will set the fireworks to be detected. It will place the fireworks at the farthest distance in the recognition area and keep the camera lens in a shallow depth of field. In this way, when the fireworks are burning, the size of the blurred light spot in the camera image can be accurately recorded. This size will be used as a threshold to determine whether the blurred light spot is a pending light spot.

[0056] Once the set range and set threshold are determined, the system will process the pending light spots. It will calculate the minimum circumscribed rectangular area of ​​each pending light spot and mark the area as the target area. This target area will be used in the subsequent image recognition and alarm information generation process. In this way, the system can more accurately locate potential fireworks targets, thereby improving the accuracy and reliability of recognition.

[0057] 4. When the system identifies a potential firework target, it will further adjust the camera lens depth of field to a deep depth of field to ensure that the entire ROI recognition area is within the imaging depth of field of the camera lens. This step is crucial to improving the accuracy of image recognition, because only when the entire ROI area is clearly visible can the image recognition model accurately analyze and judge.

[0058] In order to achieve a deep depth of field effect, the system adjusts the aperture value of the camera lens. The size of the aperture value directly affects the depth of field range of the camera. A smaller aperture value can increase the depth of field, allowing more scene elements to remain in clear focus. By precisely controlling the aperture value, the system can ensure that all details in the ROI area are captured clearly.

[0059] After adjusting the depth of field of the camera lens, the system will collect image frames of the ROI identification area again. This time, due to the deep depth of field setting, the resulting image will be clearer and more detailed. Next, the system will process the collected image, especially for the target area marked previously. The system will crop out the image block corresponding to the target area, which contains information about the possible fireworks target and its surrounding environment.

[0060] The system will then input this image block into a preset image recognition model. This model is specially trained to recognize various types of fireworks features. Deep learning models, such as convolutional neural networks, long short-term memory networks, and transformer models, are introduced for feature extraction and classification recognition of video frames. By training these models, they can extract key features from video frames and accurately detect and recognize targets. The model will analyze the input image block and output a recognition result. If the recognition result is fireworks, the system will immediately trigger the alarm push mechanism and send alarm information to relevant personnel, including:

[0061] Alarm monitoring: Various indicators, states and data of structured data generated by algorithm analysis will immediately trigger alarm push once abnormal or unexpected behavior is found.

[0062] Alarm notification: After the alarm is triggered, notifications are sent to designated users in a variety of ways, such as SMS, email, instant messaging and other tools, based on the preset alarm strategy.

[0063] Alarm handling: Handle the alarms generated by algorithm analysis, including confirming alarms, ignoring alarms, resolving alarms, and setting corresponding handling.

[0064] Alarm configuration strategy: allows users to customize alarm strategies and configurations, including alarm templates, alarm levels, and alarm strategy settings. You can set a customized alarm template to send alarm notifications. You can set a customized alarm template to send alarm notifications. You can set a customized alarm template to send alarm notifications.

[0065] If the recognition result is not fireworks, the system will continue to perform spot detection on the ROI recognition area. This means that the system has not given up monitoring potential fireworks targets, but continues to detect and analyze to ensure that no possible safety hazards are missed. This continuous monitoring method helps to improve the reliability and accuracy of the system and provide users with greater security.

[0066] The present invention also includes a proactive warning information generation system based on AI video analysis, which is used to implement the proactive warning information generation method based on AI video analysis, including:

[0067] The task configuration module is used to create a fire and smoke detection task, set the task execution time, select the corresponding video analysis algorithm in the algorithm list and draw the ROI recognition area; set the corresponding video frame extraction frequency;

[0068] A depth of field adjustment module is used to adjust the depth of field of the camera lens to a shallow depth of field or a deep depth of field;

[0069] The light spot monitoring module is used to mark the blurred light spots of known light sources in the ROI identification area, and perform real-time light spot detection on the ROI identification area according to the video frame extraction frequency; when an unmarked blurred light spot appears in the ROI identification area, the color temperature value and the maximum diameter of the blurred light spot are identified, and the blurred light spot with a color temperature value within a set range and a maximum diameter less than a set threshold is marked as a pending light spot, and the minimum circumscribed rectangular area of ​​the pending light spot is marked as the target area;

[0070] The image recognition module is used to collect the image frame of the ROI recognition area, cut the image block corresponding to the target area, input the image block into a preset image recognition model, and output the recognition result of the image block.

[0071] The alarm prompt module is used to trigger an alarm push when the recognition result is fireworks, otherwise continue to perform spot detection on the ROI recognition area.

[0072] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0073] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0074] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0075] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for generating active warning information based on AI video analysis, characterized in that: The following steps are involved: Select a camera from the camera list as the monitoring object, create a fire and smoke detection task, set the task execution time, select the corresponding video analysis algorithm in the algorithm list and draw the ROI recognition area; set the corresponding video frame rate; Adjust the depth of field of the camera lens to a shallow depth of field to blur the ROI recognition area, mark the blurred light spots of known light sources in the ROI recognition area, and perform real-time light spot detection on the ROI recognition area according to the video frame extraction frequency; When an unmarked blurred light spot appears in the ROI identification area, the color temperature value and the maximum diameter of the blurred light spot are identified, and the blurred light spot with a color temperature value within the set range and a maximum diameter less than the set threshold is marked as a pending light spot, and the minimum circumscribed rectangular area of ​​the pending light spot is marked as the target area; Adjust the camera lens depth of field to deep depth of field, collect the ROI recognition area image frame again, crop the image block corresponding to the target area, input the image block into a preset image recognition model, output the recognition result of the image block, and trigger an alarm push when the recognition result is fireworks, otherwise continue to perform spot detection on the ROI recognition area.

2. According to claim 1, a method for generating active warning information based on AI video analysis is characterized in that: The process of adjusting the depth of field of the camera lens to a shallow depth of field is: The distance between the focal plane of the camera and the camera lens is adjusted so that the distance is within the interval [A, A+a], where A is the minimum focusing distance of the camera and a is the set redundancy value.

3. The method for generating active warning information based on AI video analysis according to claim 1, characterized in that: The process of adjusting the camera lens depth of field to deep depth of field is: Adjust the aperture value of the camera lens so that the entire ROI recognition area is within the imaging depth of field of the camera lens.

4. The method for generating active warning information based on AI video analysis according to claim 1, characterized in that: The process of real-time spot detection of the ROI recognition area is as follows: A selection box is set according to the shape of the lens aperture blades, and the deformation range of the selection box is set. The image frame in the ROI recognition area is edge detected and binarized to obtain a binary image. The marked blurred light spot is excluded from the binary image. The contour of the connected area with a gray value of 255 in the binary area is selected based on the selection box, and the contour within the deformation range of the selection box is screened out and marked as the light spot contour. The light spot contour and the internal area are the blurred light spot.

5. The method for generating active warning information based on AI video analysis according to claim 1, characterized in that: The process of identifying the color temperature value of the blurred light spot is: For any blurred light spot, when the color of the blurred light spot is white, the color temperature of the blurred light spot is the white balance parameter of the current camera; When the color of the blurred light spot is light blue or blue, gradually increase the white balance parameter of the camera until the color of the blurred light spot turns white. The color temperature of the blurred light spot is the white balance parameter of the current camera. When the color of the blurred light spot is yellow or orange, the white balance parameter of the camera is gradually reduced until the color of the blurred light spot becomes white, and the color temperature of the blurred light spot is the white balance parameter of the current camera.

6. The method for generating active warning information based on AI video analysis according to claim 1, characterized in that: The process of identifying the maximum diameter of the blurred light spot is: For any blurred light spot, select several points in the outline of the blurred light spot at equal intervals, take each point as the starting point, draw a connecting line between the starting point and the other side of the outline through the centroid of the outline, and select the longest connecting line to mark it as the maximum diameter of the blurred light spot.

7. The method for generating active warning information based on AI video analysis according to claim 1, characterized in that: The specific process of marking the blurred light spot whose color temperature value is within the set range and whose maximum diameter is less than the set threshold as the pending light spot is as follows: Obtain the flame color temperature K of the fireworks to be detected, set the error threshold k, and then the setting range is [Kk, K+k]; The fireworks to be detected are set at the farthest distance in the identification area, the camera lens is kept at a shallow depth of field, and the blurred spot size of the fireworks in the camera image is recorded at this time, and this size is used as the set threshold.

8. A proactive warning information generation system based on AI video analysis, used to implement a proactive warning information generation method based on AI video analysis as described in any one of claims 1 to 7, characterized in that: include: The task configuration module is used to create a fire and smoke detection task, set the task execution time, select the corresponding video analysis algorithm in the algorithm list, and draw the ROI recognition area; Set the corresponding video frame rate; A depth of field adjustment module is used to adjust the depth of field of the camera lens to a shallow depth of field or a deep depth of field; The light spot monitoring module is used to mark the blurred light spots of known light sources in the ROI identification area, and perform real-time light spot detection on the ROI identification area according to the video frame extraction frequency; when an unmarked blurred light spot appears in the ROI identification area, the color temperature value and the maximum diameter of the blurred light spot are identified, and the blurred light spot with a color temperature value within a set range and a maximum diameter less than a set threshold is marked as a pending light spot, and the minimum circumscribed rectangular area of ​​the pending light spot is marked as the target area; The image recognition module is used to collect the image frame of the ROI recognition area, cut the image block corresponding to the target area, input the image block into a preset image recognition model, and output the recognition result of the image block. The alarm prompt module is used to trigger an alarm push when the recognition result is fireworks, otherwise continue to perform spot detection on the ROI recognition area.

Images