Large-scale outdoor monitoring camera intelligent calibration method and calibration device based on double cameras

Abstract

The invention discloses a large-scale outdoor monitoring camera intelligent calibration method and calibration device based on double cameras, and the calibration method comprises the steps: enabling a calibration device with a binocular camera to move in a space where each outdoor monitoring camera is located, and enabling the binocular camera and the outdoor monitoring camera to carry out the mutual shooting; matching the image data shot by the binocular camera with the pose information of the binocular camera through a timestamp; determining internal parameters of the outdoor monitoring camera through a deep learning network according to image data shot by the outdoor monitoring camera; identifying two-dimensional image coordinates of homonymy points in the image shot by the binocular camera and the image shot by the outdoor monitoring camera, and determining a control point; based on the two-dimensional image coordinates of the control points and the internal and external parameters of the binocular camera, calculating the three-dimensional geographic space coordinates of the control points through a triangulation algorithm; and quickly resolving pose information of the outdoor monitoring camera according to the two-dimensional image coordinates and the three-dimensional geographic space coordinates of the control points and internal parameters of the outdoor monitoring camera.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-camera system calibration, and particularly to an intelligent calibration method and calibration device for large-scale outdoor surveillance cameras based on dual cameras. Background Art

[0002] With the development of urbanization, the number of surveillance cameras has increased rapidly in a short period. However, when installing surveillance cameras, information such as geographical location and field of view coverage is often lacking, which affects effective camera layout management and spatial analysis applications. This not only makes the management and maintenance of surveillance cameras difficult, but also causes problems such as resource redundancy and low utilization rate.

[0003] Camera calibration is an important method to solve the lack of pose information of current urban surveillance cameras. Among the existing methods, one type is traditional camera calibration methods, including camera calibration methods based on markers, camera calibration methods based on active vision, and self-calibration methods, etc. These methods often have high accuracy, but they only take single-camera calibration as their main goal and do not have sufficient efficiency and universality. Therefore, when facing the calibration task of large-scale urban outdoor surveillance cameras, it is difficult to actually carry out applications. Another type of method is the census method for large-scale cameras, including the census method based on street view image detectors, camera census based on the BSP principle, etc. This type of method improves the calibration efficiency and reduces the calibration cost, but there are still problems such as incomplete solution information and being affected by the time synchronization error of the monitoring system, and it cannot fully and effectively support the rapid calibration task of large-scale urban surveillance cameras. Summary of the Invention

[0004] Object of the Invention: The present invention aims to provide an intelligent and rapid calibration method for large-scale outdoor surveillance cameras based on a calibration device with a binocular camera, and another object of the present invention is to provide an intelligent calibration device for large-scale outdoor surveillance cameras based on dual cameras.

[0005] Technical Solution: The intelligent calibration method for large-scale outdoor surveillance cameras based on dual cameras according to the present invention includes the following steps:

[0006] (1) A calibration device with a binocular camera moves in the space where each outdoor surveillance camera is located. During the movement, the binocular camera takes pictures of the outdoor surveillance camera and records the pose information of the binocular camera. At the same time, the outdoor surveillance camera also takes pictures of the calibration device with the binocular camera; match the image data taken by the binocular camera with the pose information of the binocular camera through time stamps;

[0007] (2) Determine the internal parameters of the outdoor surveillance camera through a deep learning network according to the image data taken by the outdoor surveillance camera;

[0008] (3) Identify the two-dimensional image coordinates of homologous points in the images captured by the binocular camera and the outdoor surveillance camera, and determine the control points;

[0009] (4) Based on the two-dimensional image coordinates of the control points and the internal and external parameters of the binocular camera, solve the three-dimensional geospatial coordinates of the control points through the triangulation algorithm;

[0010] (5) According to the two-dimensional image coordinates and three-dimensional geospatial coordinates of the control points and the internal parameters of the outdoor surveillance camera, use the Rance_PnP algorithm to solve the pose information of the outdoor surveillance camera.

[0011] Further, step (1) is specifically as follows:

[0012] (11) The calibration device with the binocular camera moves in the space where each outdoor surveillance camera is located. During the movement, the binocular camera takes pictures of the outdoor surveillance camera, and at the same time, the outdoor surveillance camera also takes pictures of the calibration device with the binocular camera;

[0013] (12) Based on the Kalman filter, complete the joint solution of the inertial navigation and GNSS in the pose module of the calibration device to obtain the pose data during the movement of the calibration device;

[0014] (13) Convert the pose data during the movement of the calibration device obtained in step (12) in the Earth-centered Earth-fixed coordinate system into the pose data during the movement of the calibration device in the local coordinate system;

[0015] (14) According to the time corresponding to the timestamp, combined with the relative pose between the main antenna of the pose module of the calibration device and the two cameras, solve the pose information when the binocular camera takes pictures of the corresponding time image, and match the image data captured by the binocular camera with the pose information of the binocular camera through the timestamp.

[0016] Further, step (2) is specifically as follows:

[0017] When the calibration device enters the shooting range of the outdoor surveillance camera, the outdoor surveillance camera takes pictures of the calibration device with the binocular camera, and inputs the images captured by the outdoor surveillance camera into the deep learning network to determine the internal parameters of the outdoor surveillance camera. The internal parameters include the focal length, the principal point of the image, and the distortion coefficient.

[0018] Further, step (3) is specifically as follows:

[0019] (31) Combine the left camera image and the right camera image captured by the binocular camera of the calibration device and the monitoring camera image captured by the outdoor monitoring camera in pairs to obtain the combination of the left camera image and the right camera image, the combination of the right camera image and the monitoring camera image, and the combination of the right camera image and the monitoring camera image. Input these three combinations into the SuperPoint deep learning network respectively to obtain the image feature points and description factors corresponding to the two sets of images in the combination.

[0020] (32) Input the image feature points and description factors corresponding to the two sets of images in the combination into the LightGlue deep learning network to obtain the two-dimensional image coordinates of the corresponding points and the matching confidence of a set of image feature points.

[0021] (33) Remove the corresponding points with a matching confidence less than the set threshold; set the corresponding points that exist in all three combinations as control points.

[0022] Further, step (4) is specifically as follows:

[0023] (41) Obtain the internal parameters of the binocular camera of the calibration device through the Zhang Zhengyou calibration method.

[0024] (42) According to the pose information of the left and right cameras of the binocular camera, calculate the relative external parameters between the left and right cameras of the calibration device.

[0025] (43) Based on the two-dimensional image coordinates of the control points and the internal and external parameters of the binocular camera, calculate the three-dimensional geospatial coordinates of the control points through the triangulation algorithm.

[0026] The intelligent calibration device for large-scale outdoor monitoring cameras based on dual cameras of the present invention includes

[0027] A pose module for obtaining the pose data of the calibration device when moving along the space where each monitoring device is located;

[0028] A camera module for obtaining the dual-camera image data of the calibration device when moving along the space where each monitoring device is located;

[0029] A device main body, fixedly connected to the pose module and the camera module, serving as a carrier to install the device on a mobile vehicle;

[0030] A data acquisition module for obtaining the monitoring image data captured by each outdoor monitoring camera when the calibration device moves along the space where each monitoring device is located;

[0031] An internal parameter identification module for the outdoor monitoring camera, used to identify the internal parameters of the monitoring camera;

[0032] A common corresponding point identification module for identifying the two-dimensional image coordinates of the common corresponding points between the left camera image, the right camera image and the outdoor monitoring camera image of the calibration device;

[0033] The three-dimensional coordinate calculation module of the control point is used to calculate the three-dimensional space coordinates of the control points participating in the PnP solution of the pose of the outdoor surveillance camera.

[0034] The position estimation module is used to estimate the spatial position and azimuth angle of each outdoor surveillance camera based on the two-dimensional and three-dimensional coordinates of the control points in the outdoor surveillance camera and the PnP algorithm.

[0035] Beneficial effects: Compared with the prior art, the remarkable advantages of the present invention are as follows: By moving the calibration device with a binocular camera in the outdoor space, the binocular camera takes pictures of the outdoor surveillance camera, obtains the images with the pose information of the binocular camera, and combines them with the images of the calibration device taken by the surveillance camera at a nearby moment to estimate the spatial position and attitude orientation of the surveillance camera, making the estimation of the spatial position and azimuth angle of the outdoor surveillance device faster, more efficient and less costly, and providing a new idea for the investigation of the deployment status and the optimization of the spatial layout of the outdoor surveillance camera. Description of the Drawings

[0036] Figure 1 is the flow chart of the calibration method of the present invention;

[0037] Figure 2 is the structural schematic diagram of the calibration device of the present invention;

[0038] Figure 3 is the structural diagram of the deep learning network for identifying the internal parameters of the outdoor surveillance camera;

[0039] Figure 4 is the schematic diagram of the result of homologous point identification; where (a) is the image of the outdoor surveillance camera, (b) is the image of the left camera of the calibration device, and (c) is the image of the right camera of the calibration device;

[0040] Figure 5 is the distribution map of the surveillance cameras in the experimental area;

[0041] Figure 6 is the schematic diagram of the comparison between the positioning result of the present invention and the real position. Detailed Embodiments

[0042] The following further describes the present invention with reference to the drawings.

[0043] The intelligent calibration method for large-scale outdoor surveillance cameras based on dual cameras of the present invention includes the following steps:

[0044] (1) The calibration device with a binocular camera moves in the space where each outdoor surveillance camera is located. During the movement, the binocular camera takes pictures of the outdoor surveillance camera and records the pose information of the binocular camera. At the same time, the outdoor surveillance camera also takes pictures of the calibration device with the binocular camera; the image data captured by the binocular camera is matched with the pose information of the binocular camera through timestamps.

[0045] (11) The calibration device with a binocular camera moves in the space where each outdoor surveillance camera is located. During the movement, the binocular camera takes pictures of the outdoor surveillance camera. At the same time, the outdoor surveillance camera also takes pictures of the calibration device with the binocular camera.

[0046] (12) Based on the Kalman filter, the combined solution of the inertial navigation and GNSS in the pose module of the calibration device is completed to obtain the pose data during the movement of the calibration device.

[0047] The calibration device is equipped with a positioning sensor and an inertial navigation sensor. During the movement, it will perform real-time positioning and attitude determination according to the acquisition frequency, that is, it will perform an acquisition every once in a while to obtain a set of pose data. Through recursive optimization of the Kalman filter, the absolute position information provided by the positioning sensor and the high-frequency relative motion information provided by the inertial navigation sensor are fused to output the optimal pose information of the system in the dynamic system.

[0048] Analyze the pose data and convert it to the local coordinate system to obtain the pose of the calibration device at each acquisition moment during the movement in the space where each monitoring device is located.

[0049] Among them, analyze the original positioning message data of the positioning sensor to obtain the longitude, latitude, and elevation data of the phase center of the sensor antenna at any moment, and then calculate the position coordinates in the local coordinate system. The calculation process is as follows:

[0050] First, convert the WGS84 coordinate system of the positioning data to the ECEF (Earth-Centered, Earth-Fixed coordinate system). The expression is as follows:

[0051] x = (N + H) × cos(B) × cos(L)

[0052] y = (N + H) × cos(B) × sin(L)

[0053] z = (N × (1 - e × e) + H) × sin(L)

[0054] Among them, N is the curvature radius of the reference ellipsoid, B, L, and H respectively represent the latitude, longitude, and elevation of the phase center of the antenna, and x, y, and z represent the coordinates in the converted Earth-Centered, Earth-Fixed coordinate system.

[0055] (13) Convert the pose data of the calibration device during movement obtained in the Earth-Centered Earth-Fixed (ECEF) coordinate system in step (12) into the pose data of the calibration device during movement in the local coordinate system.

[0056] The conversion from the Earth-Centered Earth-Fixed coordinate system to the local coordinate system is expressed as follows:

[0057]

[0058] where B0 and l0 are the longitude and latitude coordinates of the origin of the local coordinate system, (Δ x , Δ y , Δ z ) is the difference between the coordinates of the main antenna of the pose module of the calibration device in the Earth-Centered Earth-Fixed coordinate system and the origin of the local coordinate system, and (x, y, z) are the final position coordinates.

[0059] (14) According to the time corresponding to the timestamp, combined with the relative pose between the main antenna of the pose module of the calibration device and the dual cameras, calculate the pose information of the binocular cameras when taking images at the corresponding time, and match the image data captured by the binocular cameras with the pose information of the binocular cameras through the timestamp.

[0060] Obtain the dual-camera images collected during the movement of the dual-camera module of the calibration device along the space where each monitoring device is located and the corresponding image generation time. When the image acquisition command is executed, read the timestamp information in the pose module of the calibration device.

[0061] According to the image generation time of the dual-camera images, obtain the pose information of the pose module of the calibration device at the corresponding moment. Based on this pose information, combined with the relative pose between the main antenna of the pose module of the calibration device and the dual cameras, calculate the pose information of each dual-camera image.

[0062] Let O I -X I Y I Z I represent the combined navigation system coordinate system, O L -X L Y L Z L be the left calibration device camera coordinate system, O R -X R Y R Z R be the right calibration device camera coordinate system, then a point (X I -X I Y I Z I in the coordinate system is converted to the on-vehicle camera coordinate system (X I , Y I , Z I ) is converted to the on-vehicle camera coordinate system (X L,Y L ,Z L ) and (X R ,Y R ,Z R ) has the following equation:

[0063]

[0064] wherein, R I-L , R I-R represents the rotation matrix from the integrated navigation coordinate system to the device camera coordinate system, and t I-L , t I-R represents the translation vector from the integrated navigation coordinate system to the calibrated device camera coordinate system.

[0065] (2) According to the image data captured by the outdoor surveillance camera, determine the internal parameters of the outdoor surveillance camera through a deep learning network.

[0066] When the calibration device enters the shooting range of the outdoor surveillance camera, the outdoor surveillance camera shoots the calibration device with a binocular camera, inputs the captured image of the outdoor surveillance camera into the deep learning network, and determines the internal parameters of the outdoor surveillance camera. The internal parameters include focal length, principal point of the image, and distortion coefficient.

[0067] This step specifically includes:

[0068] (21) According to the data acquisition time of the calibration device, obtain the surveillance camera images at the adjacent moment.

[0069] Record the time when the calibration device starts to acquire data, and intercept the surveillance images of each surveillance camera at the adjacent moment through the surveillance system.

[0070] (22) Input the surveillance images into the deep learning network based on the DenseNet-161 architecture to obtain the prediction results corresponding to the focal length, principal point of the image, and distortion coefficient k 1 in the internal parameters of the surveillance camera.

[0071] Build a deep learning network based on the DenseNet-161 architecture for identifying the internal parameters of the surveillance camera. The network structure is as Figure 3 shown. Use the surveillance camera image dataset with known internal parameter labels to carry out network training, and finally realize obtaining the prediction results corresponding to the focal length, principal point of the image, and distortion coefficient k 1 in the internal parameters of the surveillance camera with a single surveillance image as the input.

[0072] (3) Identify the two-dimensional image coordinates of the homologous points in the images captured by the binocular camera and the outdoor surveillance camera, and determine the control points.

[0073] The left camera image, the right camera image, and the monitoring camera image in a set of calibration devices are respectively combined in pairs and input into the SuperPoint deep learning network to obtain image feature points and descriptors.

[0074] The output of each group of images includes: the two-dimensional coordinates (x, y) of the feature points and the descriptors (d) of the feature points.

[0075] The image feature points and descriptors are input into the LightGlue deep learning network to obtain the two-dimensional image coordinates of corresponding points and the matching confidence (from 0 to 1).

[0076] The obtained feature point coordinates (x, y) and descriptors (d) are input into the LightGlue network. Each group of inputs will be a pair of feature point sets, which are the feature point sets between the left camera image and the right camera image, the right camera image and the monitoring camera image, and the left camera image and the monitoring camera image respectively.

[0077] Calculate the two-dimensional image coordinates of the common corresponding points in the three types of images according to the two-dimensional image coordinates of the corresponding points.

[0078] For each pair of images (left - right, right - monitoring, left - monitoring), check whether there are feature points that are successfully matched and have a high matching confidence (>0.5). If the matching is successful, these feature points are regarded as corresponding points, and the recognition results are as Figure 4 shown. The common corresponding points in the three images will be the matching feature points that can be found in all three types of images, and these points will be used as control points and input into the subsequent PnP solution.

[0079] (4) Based on the two-dimensional image coordinates of the control points and the internal and external parameters of the binocular camera, solve the three-dimensional geospatial coordinates of the control points through the triangulation algorithm.

[0080] Obtain the camera internal parameters of the dual cameras of the calibration device through the Zhang Zhengyou calibration method. Through the corner detection of the checkerboard, obtain the two-dimensional point coordinates under each perspective. Then, use these two-dimensional coordinates and the known three-dimensional world coordinates to estimate the camera internal parameters by the least squares method.

[0081] According to the pose information of the left and right camera images of the calibration device, solve the relative external parameters between the left and right cameras of the calibration device. The relative rotation matrix R rel and the translation vector T rel are solved through the following formula:

[0082]

[0083] T rel = T R - R rel T L

[0084] Among them, is the transpose of the rotation matrix of the left camera, R R is the rotation matrix of the right camera, T R and T L are the translation vectors of the left and right cameras.

[0085] Based on the two-dimensional image coordinates of homologous points and the internal and external parameters of the binocular cameras, the three-dimensional geospatial coordinates of the control points are solved by the triangulation algorithm.

[0086] Among them, the projection from the camera coordinate system to the two-dimensional image coordinate system can be expressed by the following formula:

[0087]

[0088] According to this formula, the coordinates (x c1 , y c1 , z c1 ) and (x c2 , y c2 , z c2 ) in the left and right camera coordinate systems can be calculated from the two-dimensional image coordinates of homologous points and the camera internal parameters.

[0089] The process of projecting a point in the three-dimensional space coordinate system onto the camera coordinate system can be expressed by the following formula:

[0090]

[0091] According to this formula, we can calculate the corresponding relationship between the homologous points in the camera coordinate system and their coordinates in the three-dimensional space:

[0092]

[0093] By combining the above two formulas, we can respectively obtain:

[0094]

[0095] Next, by using the known relative external parameters of the left and right cameras and the homologous points a 1 , a 2 , solve for the point M(x w , y w , z w ).

[0096] Among them, the number of unknowns is 3, and the number of equations is 4. Therefore, the above two formulas are converted into the following formula:

[0097]

[0098] Use SVD decomposition to calculate an overdetermined system of equations in the form of Ax = 0. Divide the first three eigenvectors in the V matrix by the last one to obtain [X / W, Y / W, Z / W, 1], and calculate the three-dimensional coordinates (x w , y w , z w ) = (X / W, Y / W, Z / W) of the control point M.

[0099] (5) According to the two-dimensional image coordinates and three-dimensional geospatial coordinates of the control points and the internal parameters of the outdoor surveillance camera, use the Rance_PnP algorithm to solve the pose information of the outdoor surveillance camera.

[0100] Randomly select 4 point pairs from the control points, calculate a candidate camera pose solution [R, T] through the PnP algorithm, use all the point pairs, project the calculated camera pose [R, T] back to the image coordinate system, and then calculate the reprojection error where (u', v') is the projected point estimated by the current pose, and (u, v) is the true image coordinate.

[0101] According to the reprojection error, select the points with a reprojection error less than 10 pixels as inliers. Repeat the above steps until the maximum number of iterations of 50 times is satisfied, or more than 15 inliers are found. Recalculate the optimal camera pose [R, T] through all the inliers as the final solution.

[0102] The intelligent calibration device for large-scale outdoor surveillance cameras based on dual cameras described in the present invention includes

[0103] A pose module for obtaining pose data when the calibration device moves along the space where each monitoring device is located;

[0104] A camera module for obtaining dual-camera image data when the calibration device moves along the space where each monitoring device is located;

[0105] A device main body fixedly connected to the pose module and the camera module, which plays a bearing role and enables the device to be installed on a moving vehicle;

[0106] A data acquisition module for obtaining the monitoring image data captured by each outdoor surveillance camera when the calibration device moves along the space where each monitoring device is located;

[0107] An outdoor surveillance camera internal parameter identification module for identifying the internal parameters of the surveillance camera;

[0108] A common homologous point identification module for identifying the two-dimensional image coordinates of the common homologous points between the left camera image, the right camera image of the calibration device and the outdoor surveillance camera image;

[0109] The 3D coordinate calculation module of control points is used to calculate the 3D spatial coordinates of control points participating in the PnP solution of the outdoor surveillance camera pose;

[0110] The position estimation module is used to estimate the spatial position and azimuth angle of each outdoor surveillance camera based on the PnP algorithm by combining the 2D and 3D coordinates of control points in the outdoor surveillance camera.

[0111] As Figure 2 shown, the calibration device of the present invention includes a right vehicle-mounted camera 1, a GNSS antenna 3, a left vehicle-mounted camera 2, a combined navigation device 4, a power supply 5, and a vehicle fixing device 6. The calibration device main body, the pose module, and the dual-camera module are fixedly connected, and any module cannot move independently. There are no markers in the device, and the viewing directions of the two cameras are the same. Both cameras capture images directly in front of the device, and the field of view has a high degree of overlap.

[0112] For the simulation verification of the present invention, Xianlin Campus of Nanjing Normal University is selected as the experimental area. As Figure 5 shown, the area covers an area of approximately 243,018 m 2 , and there are basic geographical scene elements such as roads, shopping malls, playgrounds, teaching buildings, power distribution rooms, squares, bus stops, etc. in the area, which conform to the common characteristics of the outdoor surveillance camera layout scene. There are 319 video surveillance cameras in the experimental area, which are scattered at the intersections of internal roads and the outer facades of buildings in the area. The implementation is carried out in the Windows 10 64-bit system environment using the C++ language. Without knowing the camera positions, the positions of the north area surveillance cameras are measured using both the manual measurement method and the method proposed in this paper. The results are as Figure 6 shown. It can be seen that the present invention accurately calibrates the pose information of the surveillance cameras.

Claims

1. A large-scale outdoor surveillance camera intelligent calibration method based on dual cameras, characterized in that: The following steps are involved: (1) The calibration device with the binocular camera moves in the space where each outdoor surveillance camera is located. During the movement, the binocular camera shoots the outdoor surveillance camera and records the position and posture information of the binocular camera. At the same time, the outdoor surveillance camera also shoots the calibration device with the binocular camera. The image data captured by the binocular camera and the position and posture information of the binocular camera are matched through the timestamp; (2) Determine the internal parameters of the outdoor surveillance camera through a deep learning network based on the image data captured by the outdoor surveillance camera; (3) Identify the two-dimensional image coordinates of the same-name points in the images taken by the binocular camera and the images taken by the outdoor surveillance camera, and determine the control points; (4) Based on the two-dimensional image coordinates of the control points and the internal and external parameters of the binocular camera, the three-dimensional geographic spatial coordinates of the control points are solved by a triangulation algorithm; (5) According to the two-dimensional image coordinates and three-dimensional geographic space coordinates of the control points and the internal parameters of the outdoor surveillance camera, the Rance_PnP algorithm is used to solve the position information of the outdoor surveillance camera.

2. According to claim 1, the large-scale outdoor surveillance camera intelligent calibration method based on dual cameras is characterized in that: Step (1) is as follows: (11) The calibration device with the binocular camera moves in the space where each outdoor surveillance camera is located. During the movement, the binocular camera takes a picture of the outdoor surveillance camera, and the outdoor surveillance camera also takes a picture of the calibration device with the binocular camera. (12) Based on Kalman filtering, the joint calculation of the inertial navigation and GNSS in the posture module of the calibration device is completed to obtain the posture data of the calibration device during movement; (13) converting the position and posture data of the calibration device during the movement obtained in step (12) in the Earth-centered Earth-fixed coordinate system into the position and posture data of the calibration device during the movement in the local coordinate system; (14) According to the time corresponding to the timestamp, combined with the relative posture between the main antenna of the posture module of the calibration device and the dual cameras, the posture information of the binocular camera when taking the image at the corresponding time is solved, and the image data taken by the binocular camera is matched with the posture information of the binocular camera through the timestamp.

3. The large-scale outdoor surveillance camera intelligent calibration method based on dual cameras according to claim 1 is characterized in that: Step (2) is as follows: When the calibration device enters the shooting range of the outdoor surveillance camera, the outdoor surveillance camera shoots the calibration device with the binocular camera, and inputs the image taken by the outdoor surveillance camera into the deep learning network to determine the internal parameters of the outdoor surveillance camera. The internal parameters include focal length, image principal point and distortion coefficient.

4. According to the dual-camera large-scale outdoor surveillance camera intelligent calibration method of claim 1, it is characterized in that: Step (3) is as follows: (31) The left camera image and the right camera image taken by the binocular camera of the calibration device and the surveillance camera image taken by the outdoor surveillance camera are combined in pairs to obtain a combination of the left camera image and the right camera image, a combination of the right camera image and the surveillance camera image, and a combination of the right camera image and the surveillance camera image. These three combinations are respectively input into the SuperPoint deep learning network to obtain the image feature points and description factors corresponding to the two groups of images in the combination; (32) Input the image feature points and description factors corresponding to the two groups of images in the combination into the LightGlue deep learning network to obtain the two-dimensional image coordinates of the same-name points and the matching confidence of the corresponding set of image feature points; (33) Remove the points with the same name whose matching confidence is less than the set threshold; set the points with the same name that exist in all three combinations as control points.

5. The large-scale outdoor surveillance camera intelligent calibration method based on dual cameras according to claim 1 is characterized in that: Step (4) is as follows: (41) Obtain the internal parameters of the binocular camera of the calibration device through Zhang Zhengyou calibration method; (42) Calculating the relative external parameters between the left and right cameras of the calibration device according to the posture information of the left and right cameras of the binocular camera; (43) Based on the two-dimensional image coordinates of the control points and the internal and external parameters of the binocular camera, the three-dimensional geographic coordinates of the control points are solved through a triangulation algorithm.

6. A large-scale outdoor surveillance camera intelligent calibration device based on dual cameras, characterized in that: include A posture module is used to obtain the posture data of the calibration device when it moves along the space where each monitoring device is located; A camera module, used to obtain dual-camera image data of the calibration device when it moves along the space where each monitoring device is located; The device body is fixedly connected to the posture module and the camera module, and plays a bearing role, so that the device can be installed on the mobile carrier; A data acquisition module, used to acquire monitoring image data captured by each outdoor monitoring camera when the calibration device moves along the space where each monitoring device is located; Outdoor surveillance camera internal parameter recognition module, used to identify the surveillance camera internal parameters; A common homonymous point recognition module is used to recognize the two-dimensional image coordinates of common homonymous points between the left camera image, the right camera image and the outdoor monitoring camera image of the calibration device; The control point three-dimensional coordinate calculation module is used to calculate the three-dimensional space coordinates of the control points involved in the PnP solution of the outdoor monitoring camera posture; The position estimation module is used to estimate the spatial position and azimuth of each outdoor surveillance camera based on the PnP algorithm by combining the two-dimensional and three-dimensional coordinates of the participating control points in the outdoor surveillance camera.

Images