A method for correcting a position of an AI camera

Abstract

The application relates to the technical field of camera correction, in particular to an AI camera position correction method, which comprises the following steps: acquiring a real-time picture image currently collected by a camera and a pre-stored reference picture image, performing target recognition processing on the reference picture image to obtain reference position information of a standard reference object; mapping the reference position information to the real-time picture image to obtain superimposed position information, performing target recognition processing on the real-time picture image to obtain current position information of a current reference object. In the application, intermediate state images are continuously acquired during the adjustment process and iterative comparison is carried out until the deviation enters the allowable range, forming a closed-loop control, so that manual intervention is not needed, subjective errors in manual correction are avoided, the correction time is greatly shortened, and position deviation caused by installation deviation or mechanical drift can be compensated.

Description

Technical Field

[0001] This invention relates to the field of camera calibration technology, specifically to a method for calibrating the position of an AI camera. Background Technology

[0002] Currently, traditional methods for dealing with camera displacement faults rely solely on manual fine-tuning of the lens posture. When this fine-tuning fails to eliminate the systemic algorithmic deviation caused by the viewpoint shift, it is necessary to re-perform on-site imaging calibration, full-condition sample data collection and annotation, and full retraining and parameter adaptation of the AI ​​vision model in order to repair the detection error and restore the system to normal operation. This repair process is cumbersome, has a long implementation cycle, and lacks a standardized and rapid adaptation mechanism, resulting in long equipment downtime and low production control efficiency. This has become one of the main technical pain points for the stable operation of industrial AI vision inspection systems at present.

[0003] Furthermore, traditional methods typically consider the adjustment complete after only one adjustment, without continuously acquiring intermediate state images for iterative comparison during the adjustment process. They also lack automatic determination of whether correction is needed based on the original calibration data before correction, and fail to verify the results again after correction and compare them a second time with the original calibration data. Therefore, when a single adjustment fails to completely eliminate the deviation, automatic correction is impossible; and after long-term operation, misjudgments are easily caused by a single verification error, failing to guarantee the long-term stability and reliability of the camera position. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for correcting the position of an AI camera, comprising:

[0005] The system acquires real-time images captured by the camera and pre-saved reference images, performs target recognition processing on the reference images, and obtains the reference position information of the standard reference object.

[0006] The reference position information is mapped onto the real-time image to obtain superimposed position information. Target recognition processing is then performed on the real-time image to obtain the current position information of the current reference object.

[0007] Calculate position deviation information based on the superimposed position information and the current position information, generate a camera adjustment command based on the position deviation information, and send the camera adjustment command to the camera driving device to control the camera to perform physical position adjustment;

[0008] The intermediate state image of the camera during the adjustment process is acquired, and the intermediate state image is compared with the reference position information. If the comparison result shows that the deviation still exceeds the allowable range, the current position information is updated using the intermediate state image and the step of calculating the position deviation information based on the superimposed position information and the current position information is returned. If the comparison result shows that the deviation is within the allowable range, a correction pause signal is generated.

[0009] The adjusted real-time image is acquired, and the adjusted real-time image is compared with the reference position information to obtain the correction verification result. When the correction verification result meets the preset threshold, the camera correction is determined to be complete.

[0010] Preferably, the standard reference object is a standard pallet from the production site; target recognition processing is performed on the reference image to obtain the reference position information of the standard reference object, including:

[0011] The baseline image is identified using a pre-trained target detection model to determine the bounding rectangle of the standard tray.

[0012] The center coordinates and side lengths are calculated based on the vertex coordinates of the circumscribed rectangle, and the center coordinates and side lengths together constitute the reference position information.

[0013] Preferably, the position deviation information includes horizontal deviation values ​​and vertical deviation values; calculating the position deviation information based on the superimposed position information and the current position information includes:

[0014] Compare the coordinates of the standard reference center in the superimposed position information with the coordinates of the current reference center in the current position information;

[0015] The pixel difference between the two center coordinates in the horizontal direction is calculated as the horizontal deviation value, and the pixel difference between the two center coordinates in the vertical direction is calculated as the vertical deviation value.

[0016] Preferably, generating camera adjustment instructions based on the position deviation information includes:

[0017] Based on the preset conversion relationship between pixels and actual distance, the horizontal deviation value and the vertical deviation value are converted into actual distance deviation;

[0018] Based on the current installation height and focal length parameters of the camera, calculate the required adjustment amount for each of the XYZ axes;

[0019] The adjustment amounts in the three axes are encapsulated as the camera adjustment commands.

[0020] Preferably, after sending the camera adjustment command to the camera driver, the method further includes:

[0021] Acquire intermediate images of the camera during the adjustment process;

[0022] The intermediate state image is compared with the reference position information. If the comparison result shows that the deviation still exceeds the allowable range, the adjustment command is executed and the current position information is updated using the intermediate state image. If the comparison result shows that the deviation is within the allowable range, the adjustment is stopped and a correction pause signal is output.

[0023] Preferably, before acquiring the real-time image currently captured by the camera, the method further includes:

[0024] The camera is controlled to acquire an initial image of the standard reference object according to the original calibration parameters, and the position coordinates of the standard reference object in the initial image are stored as the original calibration data.

[0025] If a deviation is detected between the current installation position of the camera and the position in the original calibration data, the current visual monitoring task is stopped, and the camera is controlled to enter the position correction mode.

[0026] Preferably, after calculating the position deviation information based on the superimposed position information and the current position information, the method further includes:

[0027] Using a marking tool, the outline of the standard reference object in the reference image is marked on the display terminal to generate marked area data;

[0028] Comparing the intermediate state image with the reference position information includes comparing the actual position of the current reference object in the intermediate state image with the marked area data.

[0029] Preferably, controlling the camera to enter position correction mode includes:

[0030] Control the camera to stop the current image acquisition task and save the currently acquired image data;

[0031] Send a position correction request to the monitoring system, and after receiving a correction confirmation signal, control the displacement actuator to unlock the camera's locking device;

[0032] Control the camera's pan-tilt unit to enter a free adjustment state in order to perform position correction operations.

[0033] Preferably, the method further includes:

[0034] After confirming that the camera calibration is complete, save the calibrated camera position parameters and resume the camera's visual monitoring task;

[0035] Accordingly, after confirming that the camera calibration is complete, the calibrated camera position parameters are saved and the camera's visual monitoring task is resumed, including:

[0036] Retrieve the original calibration data and compare the corrected camera position parameters with the original calibration data;

[0037] If the comparison results show that the deviation between the corrected camera position parameters and the original calibration data is within the allowable range, then the corrected camera position parameters are saved and the visual monitoring task is resumed.

[0038] If the comparison results show that the deviation between the corrected camera position parameters and the original calibration data exceeds the allowable range, then the step of generating camera adjustment instructions based on the position deviation information is repeated.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] (1) This invention compares the position information of the standard reference object in the reference image with the position information of the current reference object in the real-time image, calculates the pixel-level deviation and converts it into the actual physical adjustment amount, generates instructions to drive the camera to make precise physical position adjustment, continuously acquires intermediate state images and iteratively compares them until the deviation enters the allowable range, forming a closed-loop control, without manual intervention, avoiding the subjective error of manual correction, greatly shortening the correction time, and at the same time being able to compensate for the position offset caused by installation deviation or mechanical drift.

[0041] (2) This invention automatically determines whether to enter the calibration mode by comparing the original calibration data; during calibration, the intermediate state image is compared with the reference position information one by one to ensure that the position is gradually approached; after calibration, the adjusted real-time image is verified again with the reference position information, and the calibration result is compared with the original calibration data a second time. If the deviation exceeds the tolerance, the adjustment is re-executed. This mechanism effectively ensures the stability and consistency of the calibration results, avoids misjudgment that may be caused by a single verification, and improves the reliability of the camera's long-term operation. Attached Figure Description

[0042] Figure 1 This is a schematic flowchart of the overall method in one embodiment of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figure 1 This invention provides a technical solution: a method for correcting the position of an AI camera, comprising:

[0045] S1. Acquire the real-time image captured by the camera and the pre-saved reference image, perform target recognition processing on the reference image, and obtain the reference position information of the standard reference object.

[0046] S2. Map the reference position information onto the real-time image to obtain the superimposed position information. Perform target recognition processing on the real-time image to obtain the current position information of the current reference object.

[0047] S3. Calculate the position deviation information based on the superimposed position information and the current position information, generate a camera adjustment command based on the position deviation information, and send the camera adjustment command to the camera driver to control the camera to perform physical position adjustment.

[0048] S4. Acquire the intermediate state image of the camera during the adjustment process, compare the intermediate state image with the reference position information. If the comparison result shows that the deviation still exceeds the allowable range, update the current position information using the intermediate state image and return to the step of calculating the position deviation information based on the superimposed position information and the current position information. If the comparison result shows that the deviation is within the allowable range, generate a correction pause signal.

[0049] S5. Acquire the adjusted real-time image, compare the adjusted real-time image with the reference position information to obtain the correction verification result, and determine that the camera correction is complete when the correction verification result meets the preset threshold.

[0050] It should be noted that the real-time image captured by the camera is obtained, and a pre-saved reference image is retrieved. The reference image contains a standard chessboard reference object. Target recognition is performed on the image, and the coordinates of the four corner points of the chessboard are detected as reference position information.

[0051] The reference position information is mapped onto the real-time image, that is, the theoretical positions of the chessboard corner points are drawn on the real-time image; at the same time, target recognition is performed on the real-time image to detect the coordinates of the actual chessboard corner points in the current image as the current position information; the positional deviation between the two is calculated, for example, a horizontal offset of 5 pixels and a vertical offset of 3 pixels; based on this deviation, camera adjustment commands are generated, such as requiring the drive device to rotate 0.5 degrees to the left and 0.3 degrees downward, and sent to the camera gimbal motor;

[0052] During the adjustment process, intermediate state images are continuously acquired. After each adjustment step, the positions of the checkerboard corner points in the intermediate image are compared with the reference positions. If the deviation still exceeds the allowable range, such as being 2 pixels off, the current position information is updated using the intermediate image, the deviation is recalculated, and the adjustment continues. Until a comparison shows that the deviation is within the allowable range, such as being less than 1 pixel, a correction pause signal is generated, and the physical adjustment is stopped.

[0053] The adjusted real-time image is acquired and compared with the reference position information again to obtain the correction verification result, such as a maximum deviation of 0.3 pixels. When the verification result meets the preset threshold of 0.5 pixels, the camera correction is confirmed to be complete and a correction success flag is output. The whole process realizes automatic closed-loop correction of the camera position, ensuring that the monitoring screen is aligned with the reference template.

[0054] In an optional embodiment, the standard reference object is a standard pallet on the production site; target recognition processing is performed on the reference image to obtain the reference position information of the standard reference object, including:

[0055] A pre-trained object detection model is used to identify the baseline image and determine the bounding rectangle of the standard tray.

[0056] The center coordinates and side lengths are calculated based on the vertex coordinates of the circumscribed rectangle. The center coordinates and side lengths together constitute the reference position information.

[0057] It should be noted that the reference image contains a standard tray. A pre-trained YOLOv5 object detection model is loaded, inference is performed on the image, and the bounding rectangle of the standard tray is output. The top-left corner pixel coordinates are 120, 180, and the bottom-right corner coordinates are 320, 380. The center x-coordinate of the rectangle is calculated as 120 + 320 divided by 2 = 220, and the y-coordinate is 180 + 380 divided by 2 = 280. The width is 320 - 120 = 200 pixels, and the height is 380 - 180 = 200 pixels. The reference position information includes the center coordinates 220, 280 and the side lengths 200, 200, which are used for subsequent deviation calculation and correction.

[0058] In an optional embodiment, the position deviation information includes a horizontal deviation value and a vertical deviation value; calculating the position deviation information based on the superimposed position information and the current position information includes:

[0059] Compare the coordinates of the center of the standard reference object in the overlaid position information with the coordinates of the center of the current reference object in the current position information;

[0060] Calculate the pixel difference between the two center coordinates in the horizontal direction as the horizontal deviation value, and calculate the pixel difference between the two center coordinates in the vertical direction as the vertical deviation value.

[0061] It should be noted that during the camera calibration process, the coordinates of the standard reference center in the superimposed position information are 220 x and 280 y; the coordinates of the current reference center detected in the current position information are 225 x and 283 y; the horizontal pixel difference is 225 minus 220 equals 5, and the vertical difference is 283 minus 280 equals 3; therefore, the horizontal deviation is 5 pixels and the vertical deviation is 3 pixels; based on this, an adjustment command is generated, requiring the camera to move 5 pixels to the left and 3 pixels down.

[0062] In an optional embodiment, generating camera adjustment instructions based on position deviation information includes:

[0063] Based on the preset conversion relationship between pixels and actual distance, the horizontal and vertical deviation values ​​are converted into actual distance deviations.

[0064] Based on the current installation height and focal length parameters of the camera, calculate the required adjustment amount for each of the XYZ axes;

[0065] The adjustment values ​​in the three axes are encapsulated as camera adjustment commands.

[0066] It should be noted that during camera calibration, the horizontal deviation is 5 pixels and the vertical deviation is 3 pixels. Given that the conversion between pixels and actual distance is 0.1 mm per pixel, the actual horizontal distance deviation is 5 x 0.1 = 0.5 mm, and the actual vertical distance deviation is 3 x 0.1 = 0.3 mm. The current camera installation height is 2 meters, and the focal length is 8 mm. According to the pinhole imaging principle, the relationship between the actual deviation and the adjustment angle must consider distance. Calculations show that the required adjustment angle in the horizontal direction is 0.05 degrees, and in the vertical direction, it is 0.03 degrees. In addition, the camera mount has one axial rotational degree of freedom. By analyzing the angle between the reference object's edge and the horizontal line in the current image, the rotational adjustment amount is found to be 0.02 degrees. These three axial adjustment amounts are encapsulated into a JSON-formatted camera adjustment command, including the horizontal and vertical angles of the translation axis and the angle of the rotation axis, and sent to the camera gimbal drive to perform physical position adjustment.

[0067] In an optional embodiment, after sending the camera adjustment command to the camera driver, the method further includes:

[0068] Acquire intermediate images of the camera during the adjustment process;

[0069] The intermediate state image is compared with the reference position information. If the comparison result shows that the deviation still exceeds the allowable range, the adjustment command is continued and the current position information is updated using the intermediate state image. If the comparison result shows that the deviation is within the allowable range, the adjustment is stopped and a correction pause signal is output.

[0070] It should be noted that after sending the camera adjustment command to the camera driver, intermediate state images of the camera are continuously acquired during the adjustment process. Each intermediate state image is compared with the reference position information to calculate the pixel deviation between the current reference center and the reference center. If the deviation still exceeds the preset allowable range, such as a horizontal or vertical deviation greater than 1 pixel, the adjustment command is continued to be executed, and the current position information is updated using the current intermediate state image as the starting point for the next adjustment. If a comparison result shows that the deviation is within the allowable range, the adjustment is stopped immediately, and a correction pause signal is output, indicating that the physical position adjustment stage has been completed. This process realizes closed-loop feedback, ensuring that the camera gradually approaches the ideal position.

[0071] In an optional embodiment, before acquiring the real-time image currently captured by the camera, the method further includes:

[0072] The camera is controlled to acquire an initial image of the standard reference object according to the original calibration parameters, and the position coordinates of the standard reference object in the initial image are stored as the original calibration data.

[0073] If a deviation is detected between the current installation position of the camera and the position in the original calibration data, the current visual monitoring task will be stopped and the camera will be controlled to enter the position correction mode.

[0074] It should be noted that before acquiring the real-time image captured by the camera, the camera is first controlled to acquire an initial image of a standard reference object, i.e., a standard tray, according to its original calibration parameters. For example, when the camera is initially installed and debugged, a clear image of the tray is taken, and the outer rectangle of the tray is obtained through target detection. Its center coordinates and side length are stored as the original calibration data. Each time the camera is powered on again or at regular intervals, the current installation position of the camera is checked for deviation from the position in the original calibration data. The position of the tray in the current real-time image is compared with the coordinates in the original calibration data. If the difference exceeds a preset threshold, such as 2 pixels, a deviation is determined. At this time, the currently executing visual monitoring task is immediately stopped, and the camera is controlled to enter the position correction mode. The correction steps, such as acquiring the real-time image, calculating the deviation, and generating adjustment instructions, are then executed to avoid erroneous monitoring data affecting subsequent operations.

[0075] In an optional embodiment, after calculating the position deviation information based on the superimposed position information and the current position information, the method further includes:

[0076] Using a marking tool, the outline of a standard reference object in the reference image is marked on the display terminal to generate marked area data;

[0077] The intermediate state image is compared with the reference position information, including comparing the actual position of the current reference object in the intermediate state image with the marked area data.

[0078] It should be noted that after calculating the positional deviation information, the outline of the standard reference object in the reference image can be manually or automatically marked on the display terminal using a marking tool to generate marked area data. The marked area data, such as a polygon or rectangle, accurately describes the ideal position and shape of the reference object in the reference image. When comparing the intermediate state image with the reference position information, instead of simply comparing the center point, the actual position of the current reference object in the intermediate state image is matched pixel by pixel or contour with this marked area data. For example, the overlap or average distance of the contours is calculated to more accurately determine whether the deviation is within the allowable range. This method is particularly suitable for scenarios where the reference object has an irregular shape or requires alignment of fine edges. The standard reference object is a special tray with a notch. The operator clicks around the edge of the tray in the reference image on the display terminal with the mouse to generate a closed contour marked area data. During the adjustment process, an intermediate state image is acquired, the contour of the current tray is extracted through object detection, and then the average Hausdorff distance between the current contour and the marked area contour is calculated. When this distance is less than 1 pixel, the deviation is considered to be within the allowable range. This method is more reliable than simply comparing the center point and can ensure the alignment of the entire tray area.

[0079] In an optional embodiment, controlling the camera to enter a position correction mode includes:

[0080] Control the camera to stop the current image acquisition task and save the currently acquired image data;

[0081] Send a position correction request to the monitoring system, and after receiving a correction confirmation signal, control the displacement actuator to unlock the camera's locking device;

[0082] Control the camera's pan / tilt unit to enter a free-adjustment state in order to perform position correction operations.

[0083] It should be noted that when a surveillance camera shifts due to an external impact, it needs to enter position correction mode. First, a command is sent to stop the camera's current image acquisition task and save the most recent frame to local storage as a backup. Next, a position correction request is sent to the central monitoring system, including the camera number and the current offset. After the monitoring administrator clicks "confirm" on the software interface, a correction confirmation signal is returned. Upon receiving the confirmation signal, the camera's displacement actuator, such as the clutch of a stepper motor, is activated to unlock the camera's locking device, allowing the camera's pan-tilt unit to move freely. Finally, the pan-tilt unit's damper is released, entering a free adjustment state, where the camera can be manually or automatically adjusted. Subsequently, the correction process begins, acquiring real-time images and comparing them with a reference image to generate adjustment commands to drive the pan-tilt unit until correction is complete. This process ensures that correction operations are performed safely, avoiding data loss and malfunctions.

[0084] In an optional embodiment, the method further includes:

[0085] After confirming that the camera calibration is complete, save the calibrated camera position parameters and resume the camera's visual monitoring task;

[0086] Accordingly, after confirming that the camera calibration is complete, the calibrated camera position parameters are saved and the camera's visual monitoring task is resumed, including:

[0087] Retrieve the original calibration data and compare the corrected camera position parameters with the original calibration data;

[0088] If the comparison results show that the deviation between the corrected camera position parameters and the original calibration data is within the allowable range, then save the corrected camera position parameters and resume the visual monitoring task.

[0089] If the comparison results show that the deviation between the corrected camera position parameters and the original calibration data exceeds the allowable range, then the step of generating camera adjustment instructions based on the position deviation information is repeated.

[0090] It should be noted that after confirming that the camera calibration is complete, the calibrated camera position parameters are saved and the camera's visual monitoring task is resumed. Specifically, the original calibration data is retrieved, and the calibrated camera position parameters are compared with the original calibration data, such as comparing the tray center coordinates and side lengths. If the deviation is within the allowable range, such as a center offset of less than 1 pixel or a side length error of less than 2 pixels, the calibrated parameters are saved and the monitoring task is resumed. If the deviation exceeds the allowable range, such as a center offset of 5 pixels, the calibration is deemed unqualified, and the step of generating camera adjustment instructions based on the position deviation information is re-executed for physical adjustment and verification again until the calibration result meets the requirements. This closed-loop mechanism ensures the final accuracy of the calibration.

[0091] After the camera calibration is completed, the tray center in the original calibration data is retrieved as 220, 280; the detected tray center after calibration is 221, 281, with a deviation of 1 pixel horizontally and 1 pixel vertically, and an allowable range of 2 pixels. Therefore, the new parameters are saved and monitoring is resumed; if the deviation is 225, 283, which exceeds the allowable range, the deviation is recalculated as 5 and 3 pixels, a new adjustment command is generated, and the camera is driven to move again until the deviation meets the requirements.

[0092] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for correcting the position of an AI camera, characterized in that, include: The system acquires real-time images captured by the camera and pre-saved reference images, performs target recognition processing on the reference images, and obtains the reference position information of the standard reference object. The reference position information is mapped onto the real-time image to obtain superimposed position information. Target recognition processing is then performed on the real-time image to obtain the current position information of the current reference object. Calculate position deviation information based on the superimposed position information and the current position information, generate a camera adjustment command based on the position deviation information, and send the camera adjustment command to the camera driving device to control the camera to perform physical position adjustment; The intermediate state image of the camera during the adjustment process is acquired, and the intermediate state image is compared with the reference position information. If the comparison result shows that the deviation still exceeds the allowable range, the current position information is updated using the intermediate state image and the step of calculating the position deviation information based on the superimposed position information and the current position information is returned. If the comparison result shows that the deviation is within the allowable range, a correction pause signal is generated. The adjusted real-time image is acquired, and the adjusted real-time image is compared with the reference position information to obtain the correction verification result. When the correction verification result meets the preset threshold, the camera correction is determined to be complete.

2. The method for correcting the position of an AI camera according to claim 1, characterized in that, The standard reference object is the standard pallet on the production site; The reference image is subjected to target recognition processing to obtain the reference position information of the standard reference object, including: The baseline image is identified using a pre-trained target detection model to determine the bounding rectangle of the standard tray. The center coordinates and side lengths are calculated based on the vertex coordinates of the circumscribed rectangle, and the center coordinates and side lengths together constitute the reference position information.

3. The method for correcting the position of an AI camera according to claim 2, characterized in that, The position deviation information includes horizontal deviation values ​​and vertical deviation values; calculating the position deviation information based on the superimposed position information and the current position information includes: Compare the coordinates of the standard reference center in the superimposed position information with the coordinates of the current reference center in the current position information; The pixel difference between the two center coordinates in the horizontal direction is calculated as the horizontal deviation value, and the pixel difference between the two center coordinates in the vertical direction is calculated as the vertical deviation value.

4. The method for correcting the position of an AI camera according to claim 3, characterized in that, Generate camera adjustment instructions based on the position deviation information, including: Based on the preset conversion relationship between pixels and actual distance, the horizontal deviation value and the vertical deviation value are converted into actual distance deviation; Based on the current installation height and focal length parameters of the camera, calculate the required adjustment amount for each of the XYZ axes; The adjustment amounts in the three axes are encapsulated as the camera adjustment commands.

5. The method for correcting the position of an AI camera according to claim 4, characterized in that, After sending the camera adjustment command to the camera driver, the method further includes: Acquire intermediate images of the camera during the adjustment process; The intermediate state image is compared with the reference position information. If the comparison result shows that the deviation still exceeds the allowable range, the adjustment command is executed and the current position information is updated using the intermediate state image. If the comparison result shows that the deviation is within the allowable range, the adjustment is stopped and a correction pause signal is output.

6. The method for correcting the position of an AI camera according to claim 5, characterized in that, Before acquiring the real-time image currently captured by the camera, the method further includes: The camera is controlled to acquire an initial image of the standard reference object according to the original calibration parameters, and the position coordinates of the standard reference object in the initial image are stored as the original calibration data. If a deviation is detected between the current installation position of the camera and the position in the original calibration data, the current visual monitoring task is stopped, and the camera is controlled to enter the position correction mode.

7. The method for correcting the position of an AI camera according to claim 6, characterized in that, After calculating the position deviation information based on the superimposed position information and the current position information, the method further includes: Using a marking tool, the outline of the standard reference object in the reference image is marked on the display terminal to generate marked area data; Comparing the intermediate state image with the reference position information includes comparing the actual position of the current reference object in the intermediate state image with the marked area data.

8. The method for correcting the position of an AI camera according to claim 7, characterized in that, Controlling the camera to enter position correction mode includes: Control the camera to stop the current image acquisition task and save the currently acquired image data; Send a position correction request to the monitoring system, and after receiving a correction confirmation signal, control the displacement actuator to unlock the camera's locking device; Control the camera's pan-tilt unit to enter a free adjustment state in order to perform position correction operations.

9. A method for correcting the position of an AI camera according to claim 8, characterized in that, The method further includes: After confirming that the camera calibration is complete, save the calibrated camera position parameters and resume the camera's visual monitoring task; Accordingly, after confirming that the camera calibration is complete, the calibrated camera position parameters are saved and the camera's visual monitoring task is resumed, including: Retrieve the original calibration data and compare the corrected camera position parameters with the original calibration data; If the comparison results show that the deviation between the corrected camera position parameters and the original calibration data is within the allowable range, then the corrected camera position parameters are saved and the visual monitoring task is resumed. If the comparison results show that the deviation between the corrected camera position parameters and the original calibration data exceeds the allowable range, then the step of generating camera adjustment instructions based on the position deviation information is repeated.

Images