Wireless binocular vision structure three-dimensional displacement measurement method and system

Abstract

The present application belongs to the field of optical measurement and civil engineering monitoring, and proposes a wireless binocular vision structure three-dimensional displacement measurement method and system. The high-precision time information and PPS signal output by the GNSS timing module are used to realize the wireless hardware-level synchronous trigger collection of the left and right cameras. The edge-cloud collaborative calculation is introduced: in the initialization stage, only the first frame of the left and right cameras is uploaded to the cloud to complete feature matching and issue tracking targets; in the running stage, the edge end performs light template matching and sub-pixel precision positioning, and only uploads pixel coordinates; the cloud end triangulates the coordinates combined with the projection matrix obtained by calibration, and provides real-time calculation of the three-dimensional displacement of the target and the result service. The present application avoids the wiring difficulty and data transmission bottleneck in the wide baseline scene, supports flexible adjustment of the camera baseline width, significantly reduces the bandwidth and storage overhead, improves the three-dimensional displacement measurement accuracy and system reliability in long-distance scenes, and is suitable for long-term online monitoring of large structures such as bridges, high-rise buildings and the like.

Description

Technical Field

[0001] This invention belongs to the field of optical measurement and civil engineering monitoring technology, and specifically relates to a wireless binocular vision method and system for measuring three-dimensional displacement of structures. Background Technology

[0002] Existing methods for measuring three-dimensional structural displacement commonly employ total stations, Global Navigation Satellite Systems (GNSS), and microwave interferometric radar. In recent years, with the development of computer vision, vision-based displacement measurement technology has been widely used in engineering surveying due to its advantages such as non-contact operation, low cost, and high accuracy. Since camera imaging is a three-dimensional to two-dimensional projection mapping, monocular cameras struggle to recover the depth scale in the absence of additional prior information, typically only enabling in-plane (two-dimensional) displacement measurement. To obtain three-dimensional displacement, stereo (binocular) vision measurement systems have been widely adopted.

[0003] A typical binocular vision measurement system consists of two cameras and an image processing device (such as a computer), usually connected via a wired connection for hardware triggering and data transmission. The system simultaneously acquires image sequences of the same measurement point from two viewpoints, calculates disparity through stereo matching, and recovers depth information using triangulation principles to obtain the three-dimensional displacement time history of the structure. Under fixed camera baselines (distance between the two cameras) and imaging parameters, depth measurement accuracy decreases significantly with increasing object distance; increasing the baseline width helps improve depth resolution. For long-distance scenarios such as long-span bridges or high-rise buildings, increasing the baseline is usually necessary to ensure measurement accuracy.

[0004] However, as the baseline increases, traditional wired binocular systems have revealed several problems in engineering applications: (1) long-distance cabling leads to increased hardware and construction costs and greater deployment difficulty; (2) the range of cameras that can be deployed is limited by cable length constraints; and (3) long-distance wired transmission places higher demands on synchronization triggering and data bandwidth, which can easily cause transmission bottlenecks and a decrease in system stability. These factors together restrict the widespread application of traditional wired binocular vision in large-scale three-dimensional displacement monitoring of civil structures. Summary of the Invention

[0005] The purpose of this invention is to propose a method and system for measuring three-dimensional displacement of a binocular vision structure without the need for wired signal connections. This method achieves wireless hardware-synchronized triggering of data acquisition by receiving GNSS satellite time and PPS (pulses per second) signals. Furthermore, it employs an edge-cloud collaborative strategy combining local lightweight tracking and cloud-based three-dimensional computation to avoid continuously transmitting large image sequences. This significantly reduces bandwidth and computational burden while ensuring real-time performance, thereby overcoming the technical shortcomings of existing wired binocular vision systems under wide baseline conditions, such as complex deployment, limited reliability, and high transmission overhead.

[0006] The technical solution of this invention:

[0007] A wireless binocular vision method for measuring three-dimensional displacement of a structure includes:

[0008] S1. Camera calibration and projection model establishment;

[0009] S2, GNSS timing and wireless synchronization triggering:

[0010] S3. First Frame Cloud Feature Matching and Tracking Target Determination:

[0011] S4. Local lightweight target tracking and sub-pixel precision positioning:

[0012] S5, cloud-based 3D coordinate calculation and result service.

[0013] Preferably, step S1 includes the following steps:

[0014] S11. Determine the baseline width between the left and right cameras according to the measurement requirements, and install the left and right cameras;

[0015] S12. Adjust the orientation of the left and right cameras to ensure that the structure under test is in the overlapping field of view of the left and right cameras.

[0016] S13. Use Zhang Zhengyou's calibration method to calibrate the internal parameters of the left and right cameras respectively;

[0017] Internal parameters of the left and right cameras:

[0018] (1)

[0019] A coordinate system is established on the acquired image plane, with the horizontal axis being the x-axis and the vertical axis being the y-axis, and the origin being the upper left corner; the subscripts l and r represent the left camera and the right camera, respectively; , These represent the intrinsic parameter matrices of the left and right cameras, respectively. , , , These are the focal lengths of the left and right cameras along the x and y axes, respectively, expressed in pixels. , These represent the coordinates of the principal points of the left and right cameras in the image plane, respectively.

[0020] S14. Establish a unified reference system with the left camera coordinate system as the world coordinate system. The origin is located at the optical center of the left camera, the Z-axis is along its optical axis, and the X-axis and Y-axis are parallel to the imaging plane.

[0021] S15. Measure the external parameters of the left and right cameras: the attitude and position of the left and right cameras, and the Euler angles and coordinates of the left and right cameras relative to the world coordinate system.

[0022] S16. Calculate the rotation matrix and translation vector of the left and right cameras based on the measured Euler angles and coordinates.

[0023] Let the rotation matrices of the left and right cameras relative to the world coordinate system be respectively... , The center coordinates of the left and right cameras are respectively , The relative rotation and translation of the right camera with respect to the left camera can be written as:

[0024] (2)

[0025] The world coordinate system is set as the left camera coordinate system, therefore , ;

[0026] When using Euler angles When rotating, the rotation matrices for the left and right cameras are as follows:

[0027] (3)

[0028] S17. Calculate the projection matrices of the left and right cameras using the intrinsic and extrinsic parameters of the left and right cameras.

[0029] (4)

[0030] in, , These are the projection matrices of the left and right cameras, respectively. Represents the identity matrix.

[0031] Preferably, step S2 includes the following steps:

[0032] S21. Receive GNSS satellite signals and obtain time information for atomic clock tracing;

[0033] S22. Using the standard time stamp output by the GNSS receiver in conjunction with the second pulse signal PPS, the system clock of the local computer is calibrated and synchronized via NTP or PTP.

[0034] S23. The local computer divides or reshapes the second pulse signal PPS to generate an external trigger signal for the camera, so as to obtain a synchronous trigger that matches the set frame rate.

[0035] S24. Trigger the transmission of the acquired image back to the local computer cache;

[0036] The trigger frequency should not exceed the camera's highest sampling frequency and should be matched with the camera's exposure parameters to avoid frame drops or motion blur; the camera supports external hard trigger input.

[0037] Preferably, step S3 includes the following steps:

[0038] S31. Upload the first frame images from the left and right cameras to the cloud computing platform;

[0039] S32. On a cloud computing platform, a feature point detection and matching algorithm is used to determine the corresponding feature point pairs in the left and right images obtained by the left and right cameras.

[0040] S33. Select a sub-image composed of a neighborhood with a preset size centered on a number of feature point pairs as the tracking target, and transmit the coordinates and size of each tracking target in the first frame image captured by the left and right cameras back to the local computer.

[0041] In step S3, the feature point detection and matching algorithm adopts the SIFT algorithm, SURF algorithm, or ORB algorithm; and combines the RANSAC algorithm to remove mismatches.

[0042] Preferably, step S4 includes the following steps:

[0043] S41. Based on the coordinates and size of the tracked target in the first frame image returned in step S3, extract the image region corresponding to the tracked target in the first frame images of the left camera and the right camera respectively, using the coordinates and size as the template image of the tracked target.

[0044] S42. For each frame image after the first frame image, a search window with a limited range is set in the neighborhood of the position of the tracking target in the previous frame as the center. The size of the search window is larger than the size of the template image, which is used to limit the possible position range of the tracking target in the current frame.

[0045] S43. Within the search window, a zero-mean normalized cross-correlation method is used to calculate the similarity between the template image and image regions of the same size at each candidate position within the search window; by sliding the template image pixel by pixel within the search window, the similarity coefficient between the template image and the corresponding pixel grayscale values ​​of the currently overlapping region is calculated; assuming the image within the search window is represented as... The template image is represented as The set of pixel coordinates defined in the template image is Pixel coordinates within the template image satisfy ,in and All are positive integers, representing any candidate position in the search window. ,in and All are positive integers, and the similarity coefficient is expressed as:

[0046] (5)

[0047] in, template image Average gray value, Indicates that in the search window The average grayscale value of the top-left corner region, which is the same size as the template image; by traversing all candidate positions within the search window, a similarity coefficient distribution map is obtained;

[0048] S44. Extract the location of the maximum similarity coefficient in the similarity coefficient distribution map and the similarity coefficient value within its 9×9 neighborhood;

[0049] S45. Sub-pixel interpolation and extreme value localization are performed on the extracted similarity coefficient distribution map using the Lanczos kernel. The one-dimensional Lanczos kernel is defined as:

[0050] (6)

[0051] In the formula, and These are the Lanczos nuclei along the horizontal and vertical axes, respectively. and All are real numbers; Let be the radius of the Lanczos kernel, taken as 4, i.e. The size is , The size is ; and They are respectively for and Round down; Coordinates in the similarity coefficient distribution graph The similarity coefficient at the location, and All are positive integers; The sub-pixel position similarity coefficients obtained after interpolation are... The coordinates at which the maximum value is obtained are used as the sub-pixel precision coordinates of the tracking target in the image, thereby obtaining the first... The sub-pixel precision coordinates of the tracked target in the left and right images of the frame are denoted as follows: , .

[0052] The side length of the search window in step S4 is twice the side length of the template image and is adjusted according to the movement range of the tracked target and the frame rate.

[0053] Preferably, step S5 includes the following steps:

[0054] S51, the first obtained in step S4 The sub-pixel precision coordinates of the tracked target in the left and right images of the frame are uploaded to the cloud computing platform in real time;

[0055] S52, The cloud computing platform uses the projection matrix obtained in step S1 and the first [projection matrix] uploaded in step S4. The sub-pixel precision coordinates of the tracked target in the left and right images of the frame are used to construct homogeneous constraints by linear triangulation and solve for the three-dimensional coordinates of the tracked target.

[0056] remember , The first projection matrix Okay, the construction is as follows:

[0057] (7)

[0058] Written ;right Perform singular value decomposition and take the right singular vector corresponding to the smallest singular value. Then the three-dimensional coordinates are:

[0059] (8)

[0060] in, Define the coordinates of the target point in the world coordinate system for step S14;

[0061] S53: Track the target's 3D coordinates and provide visualization and interface services in streaming or batch on the cloud computing platform, allowing users to view or subscribe in real time.

[0062] A wireless binocular vision-based three-dimensional displacement measurement system includes a left camera subsystem, a right camera subsystem, and a cloud computing module; wherein:

[0063] The left camera subsystem and the right camera subsystem have the same structure, both including:

[0064] a) Camera module, used to receive external hardware triggers and acquire images;

[0065] b) GNSS timing module, used to receive and parse GNSS time information and second pulse signal PPS to provide a local time reference;

[0066] c) The local image processing and trigger control module is configured to generate an external trigger signal with a set frame rate based on the pulse-per-second (PPS) signal to drive the camera module to acquire images synchronously, and to perform template matching and sub-pixel precision positioning on the tracked target in subsequent frames to obtain the first frame. Subpixel precision coordinates of the tracked target in the left and right images of the frame;

[0067] d) Mobile communication module, used for uplink and downlink communication with the cloud computing module;

[0068] The cloud computing module is used to perform feature point detection and matching algorithms on the first frame images captured by the left and right cameras and send down the initial coordinates of the tracking target; based on the first frame... The subpixel precision coordinates of the tracked target in the left and right images of the frame are triangulated with the left and right projection matrices to solve for the three-dimensional coordinates of the tracked target and the results are output.

[0069] The left camera subsystem and the right camera subsystem do not require a wired connection and do not need to communicate with each other.

[0070] The local image processing and trigger control module communicates with the camera module, GNSS timing module, and mobile communication module via local wired connections. The external trigger of the camera module is generated by the local image processing and trigger control module based on the GNSS pulse-per-second (PPS) signal.

[0071] The beneficial effects of this invention are:

[0072] 1. By utilizing the PPS signal and time stamp of GNSS, wireless hardware-level synchronization of binocular cameras is achieved, avoiding the difficulties in setting up wired contact devices and their susceptibility to environmental constraints. This improves deployment flexibility and system reliability, thereby enabling flexible adjustment of the baseline width according to measurement requirements and ensuring measurement accuracy in all directions during long-distance measurement applications.

[0073] 2. Only the first frame is uploaded to the cloud for feature alignment; subsequent lightweight tracking is performed locally, uploading only pixel coordinates instead of the entire frame image, which significantly reduces continuous bandwidth and storage pressure, facilitating long-term online monitoring.

[0074] 3. Improve tracking accuracy by limiting the search window, using efficient template matching algorithms and sub-pixel precision positioning (such as Lanczos / quadratic surface fitting); achieve millimeter-level three-dimensional displacement measurement capability by performing stable linear triangulation calculation based on the calibration projection matrix in the cloud (under typical civil structure line-of-sight and resolution configuration). Attached Figure Description

[0075] Figure 1 This is a schematic diagram of the equipment layout for a three-dimensional displacement measurement system for large outdoor structures according to an embodiment of the present invention;

[0076] Figure 2 This is a component of the left / right eye subsystem 1 / 2 of one embodiment of the present invention;

[0077] Figure 3 This is a flowchart of a method and system for measuring three-dimensional displacement of large outdoor structures according to an embodiment of the present invention.

[0078] In the diagram: 1. Left eye camera subsystem; 2. Right eye camera subsystem; 3. Cloud computing module; 4. Outdoor high-rise building structure; 11. Camera module; 12. GNSS timing module; 13. Local image processing and trigger control module; 14. Mobile communication module. Detailed Implementation

[0079] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.

[0080] like Figure 1 As shown, this invention provides a wireless binocular vision-based 3D structural displacement measurement system for monitoring the 3D displacement of an outdoor high-rise building structure 4. The wireless binocular vision-based 3D structural displacement measurement system includes a left-eye camera subsystem 1, a right-eye camera subsystem 2, and a cloud computing module 3. The left-eye camera subsystem 1 and the right-eye camera subsystem 2 do not require wired signal connections or mutual communication; they rely solely on their respective GNSS timing systems for cross-site time synchronization. The baseline widths of the left and right cameras are configured based on the distance to the measured structure 4 and the target accuracy.

[0081] like Figure 2 As shown, the left-eye camera subsystem 1 and the right-eye camera subsystem 2 have the same structure. Taking the left-eye camera subsystem 1 as an example, it includes:

[0082] Camera module 11 (supports external hard trigger input);

[0083] GNSS timing module 12 (outputs standard time scale and PPS signal);

[0084] Local image processing and trigger control module 13 (responsible for trigger generation, image preprocessing, target tracking and sub-pixel precision positioning);

[0085] Mobile communication module 14 (uplink / downlink data link to the cloud).

[0086] The camera module 11 is connected to the local image processing and trigger control module 13, the GNSS timing module 12 is connected to the local image processing and trigger control module 13, and the mobile communication module 14 is connected to the local image processing and trigger control module 13 via local wired connections. The cloud computing module 3 establishes data channels (public wireless network or private network) with the left eye camera subsystem 1 and the right eye camera subsystem 2 through their respective mobile communication modules.

[0087] like Figure 3 As shown, the workflow of this invention includes the following steps:

[0088] S1. Camera calibration and projection model establishment;

[0089] S2, GNSS timing and wireless synchronization triggering:

[0090] S3. First Frame Cloud Feature Matching and Tracking Target Determination:

[0091] S4. Local lightweight target tracking and sub-pixel precision positioning:

[0092] S5, cloud-based 3D coordinate calculation and result service.

[0093] Preferably, step S1 includes the following steps:

[0094] S11. Determine the baseline width between the left and right cameras according to the measurement requirements, and install the left and right cameras;

[0095] S12. Adjust the orientation of the left and right cameras to ensure that the structure under test is in the overlapping field of view of the left and right cameras.

[0096] S13. Use Zhang Zhengyou's calibration method to calibrate the internal parameters of the left and right cameras respectively;

[0097] Internal parameters of the left and right cameras:

[0098] (1)

[0099] A coordinate system is established on the acquired image plane, with the horizontal axis being the x-axis and the vertical axis being the y-axis, and the origin being the upper left corner; the subscripts l and r represent the left camera and the right camera, respectively; , These represent the intrinsic parameter matrices of the left and right cameras, respectively. , , , These are the focal lengths of the left and right cameras along the x and y axes, respectively, expressed in pixels. , These represent the coordinates of the principal points of the left and right cameras in the image plane, respectively.

[0100] S14. Establish a unified reference system with the left camera coordinate system as the world coordinate system. The origin is located at the optical center of the left camera, the Z-axis is along its optical axis, and the X-axis and Y-axis are parallel to the imaging plane.

[0101] S15. Measure the external parameters of the left and right cameras: the attitude and position of the left and right cameras, and the Euler angles and coordinates of the left and right cameras relative to the world coordinate system.

[0102] S16. Calculate the rotation matrix and translation vector of the left and right cameras based on the measured Euler angles and coordinates.

[0103] Let the rotation matrices of the left and right cameras relative to the world coordinate system be respectively... , The center coordinates of the left and right cameras are respectively , The relative rotation and translation of the right camera with respect to the left camera can be written as:

[0104] (2)

[0105] The world coordinate system is set as the left camera coordinate system, therefore , ;

[0106] When using Euler angles When rotating, the rotation matrices for the left and right cameras are as follows:

[0107] (3)

[0108] S17. Calculate the projection matrices of the left and right cameras using the intrinsic and extrinsic parameters of the left and right cameras.

[0109] (4)

[0110] in, , These are the projection matrices of the left and right cameras, respectively. Represents the identity matrix.

[0111] Preferably, step S2 includes the following steps:

[0112] S21. Receive GNSS satellite signals and obtain time information for atomic clock tracing;

[0113] S22. Using the standard time stamp output by the GNSS receiver in conjunction with the second pulse signal PPS, the system clock of the local computer is calibrated and synchronized via NTP or PTP.

[0114] S23. The local computer divides or reshapes the second pulse signal PPS to generate an external trigger signal for the camera, so as to obtain a synchronous trigger that matches the set frame rate.

[0115] S24. Trigger the transmission of the acquired image back to the local computer cache;

[0116] The trigger frequency should not exceed the camera's highest sampling frequency and should be matched with the camera's exposure parameters to avoid frame drops or motion blur; the camera supports external hard trigger input.

[0117] Preferably, step S3 includes the following steps:

[0118] S31. Upload the first frame images from the left and right cameras to a cloud computing platform (which can be a public cloud or a private cloud, such as AWS, Alibaba Cloud, Google Cloud, etc.).

[0119] S32. On a cloud computing platform, a feature point detection and matching algorithm is used to determine the corresponding feature point pairs in the left and right images obtained by the left and right cameras.

[0120] S33. Select a sub-image consisting of a neighborhood with a preset size centered on a number of feature point pairs as the tracking target (the number is determined by a trade-off between real-time performance and robustness), and transmit the coordinates and size of each tracking target in the first frame image captured by the left and right cameras back to the local computer.

[0121] Preferably, step S4 includes the following steps:

[0122] S41. Based on the coordinates and size of the tracked target in the first frame image returned in step S3, extract the image region corresponding to the tracked target in the first frame images of the left camera and the right camera respectively, using the coordinates and size as the template image of the tracked target.

[0123] S42. For each frame image after the first frame image, a search window with a limited range is set in the neighborhood of the position of the tracking target in the previous frame as the center. The size of the search window is larger than the size of the template image, which is used to limit the possible position range of the tracking target in the current frame.

[0124] S43. Within the search window, a zero-mean normalized cross-correlation method is used to calculate the similarity between the template image and image regions of the same size at each candidate position within the search window; by sliding the template image pixel by pixel within the search window, the similarity coefficient between the template image and the corresponding pixel grayscale values ​​of the currently overlapping region is calculated; assuming the image within the search window is represented as... The template image is represented as The set of pixel coordinates defined in the template image is Pixel coordinates within the template image satisfy ,in and All are positive integers, representing any candidate position in the search window. ,in and All are positive integers, and the similarity coefficient is expressed as:

[0125] (5)

[0126] in, template image Average gray value, Indicates that in the search window The average grayscale value of the top-left corner region, which is the same size as the template image; by traversing all candidate positions within the search window, a similarity coefficient distribution map is obtained;

[0127] S44. Extract the location of the maximum similarity coefficient in the similarity coefficient distribution map and the similarity coefficient value within its 9×9 neighborhood;

[0128] S45. Sub-pixel interpolation and extreme value localization are performed on the extracted similarity coefficient distribution map using the Lanczos kernel. The one-dimensional Lanczos kernel is defined as:

[0129] (6)

[0130] In the formula, and These are the Lanczos nuclei along the horizontal and vertical axes, respectively. and All are real numbers; Let be the radius of the Lanczos kernel, taken as 4, i.e. The size is , The size is ; and They are respectively for and Round down; Coordinates in the similarity coefficient distribution graph The similarity coefficient at the location, and All are positive integers; The sub-pixel position similarity coefficients obtained after interpolation are... The coordinates at which the maximum value is obtained are used as the sub-pixel precision coordinates of the tracking target in the image, thereby obtaining the first... The sub-pixel precision coordinates of the tracked target in the left and right images of the frame are denoted as follows: , .

[0131] Preferably, step S5 includes the following steps:

[0132] S51, the first obtained in step S4 The sub-pixel precision coordinates of the tracked target in the left and right images of the frame are uploaded to the cloud computing platform in real time;

[0133] S52, The cloud computing platform uses the projection matrix obtained in step S1 and the first [projection matrix] uploaded in step S4. The sub-pixel precision coordinates of the tracked target in the left and right images of the frame are used to construct homogeneous constraints by linear triangulation and solve for the three-dimensional coordinates of the tracked target.

[0134] remember , The first projection matrix Okay, the construction is as follows:

[0135] (7)

[0136] Written ;right Perform singular value decomposition and take the right singular vector corresponding to the smallest singular value. Then the three-dimensional coordinates are:

[0137] (8)

[0138] in, Define the coordinates of the target point in the world coordinate system for step S14;

[0139] S53: Track the target's 3D coordinates and provide visualization and interface services in streaming or batch on the cloud computing platform, allowing users to view or subscribe in real time.

[0140] When the target being measured is installed on the top of a 2m high pole structure, the horizontal distance between one camera module and the target is 189m and the elevation difference is 7.5m; the horizontal distance between the other camera module and the target is 195m and the elevation difference is 7.5m; and the horizontal distance between the two camera modules is 93m, the three-dimensional displacement measurement accuracy can reach 1.02mm when the measurement is carried out according to the procedure of this invention.

Claims

1. A wireless binocular vision-based three-dimensional displacement measurement method, characterized in that, include: S1. Camera calibration and projection model establishment; S2, GNSS timing and wireless synchronization triggering: S21. Receive GNSS satellite signals and obtain time information for atomic clock tracing; S22. Using the standard time stamp output by the GNSS receiver in conjunction with the second pulse signal PPS, the system clock of the local computer is calibrated and synchronized via NTP or PTP. S23. The local computer divides or reshapes the second pulse signal PPS to generate an external trigger signal for the camera, so as to obtain a synchronous trigger that matches the set frame rate. S24. Trigger the transmission of the acquired image back to the local computer cache; The trigger frequency should not exceed the camera's highest sampling frequency and should be matched with the camera's exposure parameters to avoid frame drops or motion blur; the camera supports external hard trigger input. S3. First Frame Cloud Feature Matching and Tracking Target Determination: S31. Upload the first frame images from the left and right cameras to the cloud computing platform; S32. On a cloud computing platform, a feature point detection and matching algorithm is used to determine the corresponding feature point pairs in the left and right images obtained by the left and right cameras. S33. Select a sub-image composed of a neighborhood with a preset size centered on a number of feature point pairs as the tracking target, and transmit the coordinates and size of each tracking target in the first frame image captured by the left and right cameras back to the local computer. S4. Local lightweight target tracking and sub-pixel precision positioning: S5, cloud-based 3D coordinate calculation and result service.

2. The wireless binocular vision structural three-dimensional displacement measurement method according to claim 1, characterized in that, The specific implementation process of step S1 is as follows: S11. Determine the baseline width between the left and right cameras according to the measurement requirements, and install the left and right cameras; S12. Adjust the orientation of the left and right cameras to ensure that the structure under test is in the overlapping field of view of the left and right cameras. S13. Use Zhang Zhengyou's calibration method to calibrate the internal parameters of the left and right cameras respectively; Internal parameters of the left and right cameras: (1) ； A coordinate system is established on the acquired image plane, with the horizontal axis being the x-axis and the vertical axis being the y-axis, and the origin being the upper left corner; the subscripts l and r represent the left camera and the right camera, respectively; , These represent the intrinsic parameter matrices of the left and right cameras, respectively. , , , These are the focal lengths of the left and right cameras along the x and y axes, respectively, expressed in pixels. , These represent the coordinates of the principal points of the left and right cameras in the image plane, respectively. S14. Establish a unified reference system with the left camera coordinate system as the world coordinate system. The origin is located at the optical center of the left camera, the Z-axis is along its optical axis, and the X-axis and Y-axis are parallel to the imaging plane. S15. Measure the external parameters of the left and right cameras: the attitude and position of the left and right cameras, and the Euler angles and coordinates of the left and right cameras relative to the world coordinate system. S16. Calculate the rotation matrix and translation vector of the left and right cameras based on the measured Euler angles and coordinates. Let the rotation matrices of the left and right cameras relative to the world coordinate system be respectively... , The center coordinates of the left and right cameras are respectively , The relative rotation and translation of the right camera with respect to the left camera can be written as: (2) ； The world coordinate system is set as the left camera coordinate system, therefore , ; When using Euler angles When rotating, the rotation matrices for the left and right cameras are as follows: (3) ； S17. Calculate the projection matrices of the left and right cameras using the intrinsic and extrinsic parameters of the left and right cameras. (4) ； in, , These are the projection matrices of the left and right cameras, respectively; Represents the identity matrix.

3. The wireless binocular vision structural three-dimensional displacement measurement method according to claim 1, characterized in that, In step S3, the feature point detection and matching algorithm adopts the SIFT algorithm, SURF algorithm, or ORB algorithm; and combines the RANSAC algorithm to remove mismatches.

4. The wireless binocular vision structural three-dimensional displacement measurement method according to claim 1, characterized in that, The specific implementation process of step S4 is as follows: S41. Based on the coordinates and size of the tracked target in the first frame image returned in step S3, extract the image region corresponding to the tracked target in the first frame images of the left camera and the right camera respectively, using the coordinates and size as the template image of the tracked target. S42. For each frame image after the first frame image, a search window with a limited range is set in the neighborhood of the position of the tracking target in the previous frame as the center. The size of the search window is larger than the size of the template image, which is used to limit the possible position range of the tracking target in the current frame. S43. Within the search window, a zero-mean normalized cross-correlation method is used to calculate the similarity between the template image and the image regions of the same size at each candidate position within the search window; by sliding the template image pixel by pixel within the search window, the similarity coefficient between the template image and the corresponding pixel grayscale values ​​of the current overlapping region is calculated. Assume the image within the search window is represented as The template image is represented as The set of pixel coordinates defined in the template image is Pixel coordinates within the template image satisfy ,in and All are positive integers, representing any candidate position in the search window. ,in and All are positive integers, and the similarity coefficient is expressed as: (5) ； in, template image Average gray value, Indicates that in the search window The average gray value of the area in the upper left corner that is the same size as the template image; by traversing all candidate positions within the search window, a similarity coefficient distribution map is obtained; S44. Extract the location of the maximum similarity coefficient in the similarity coefficient distribution map and the similarity coefficient value within its 9×9 neighborhood; S45. Sub-pixel interpolation and extreme value localization are performed on the extracted similarity coefficient distribution map using the Lanczos kernel. The one-dimensional Lanczos kernel is defined as: (6) ； In the formula, and These are the Lanczos nuclei along the horizontal and vertical axes, respectively. and All are real numbers; Let be the radius of the Lanczos kernel, taken as 4, i.e. The size is , The size is ; and They are respectively for and Round down; Coordinates in the similarity coefficient distribution graph The similarity coefficient at the location, and All are positive integers; The sub-pixel position similarity coefficients obtained after interpolation are... The coordinates at which the maximum value is obtained are used as the sub-pixel precision coordinates of the tracking target in the image, thereby obtaining the first... The sub-pixel precision coordinates of the tracked target in the left and right images of the frame are denoted as follows: , .

5. The wireless binocular vision structural three-dimensional displacement measurement method according to claim 4, characterized in that, The side length of the search window in step S4 is twice the side length of the template image and is adjusted according to the movement range of the tracked target and the frame rate.

6. The wireless binocular vision structural three-dimensional displacement measurement method according to claim 5, characterized in that, The specific implementation process of step S5 is as follows: S51, the first obtained in step S4 The sub-pixel precision coordinates of the tracked target in the left and right images of the frame are uploaded to the cloud computing platform in real time; S52, The cloud computing platform uses the projection matrix obtained in step S1 and the first [projection matrix] uploaded in step S4. The sub-pixel precision coordinates of the tracked target in the left and right images of the frame are used to construct homogeneous constraints by linear triangulation and solve for the three-dimensional coordinates of the tracked target. remember , The first projection matrix Okay, the construction is as follows: (7) ； Written ;right Perform singular value decomposition and take the right singular vector corresponding to the smallest singular value. Then the three-dimensional coordinates are: (8) ； in, Define the coordinates of the target point in the world coordinate system for step S14; S53: Track the target's 3D coordinates and provide visualization and interface services in streaming or batch on the cloud computing platform, allowing users to view or subscribe in real time.

7. A wireless binocular vision-based three-dimensional displacement measurement system, characterized in that, The wireless binocular vision-based structural 3D displacement measurement system includes a left camera subsystem, a right camera subsystem, and a cloud computing module; wherein: The left camera subsystem and the right camera subsystem have the same structure, both including: a) Camera module, used to receive external hardware triggers and acquire images; b) GNSS timing module, used to receive and parse GNSS time information and second pulse signal PPS to provide a local time reference; c) The local image processing and trigger control module is configured to generate an external trigger signal with a set frame rate based on the second pulse signal PPS to drive the camera module to acquire synchronously, and perform template matching and sub-pixel precision positioning on the tracked target in subsequent frames to obtain the sub-pixel precision coordinates of the tracked target in the left and right images of the nth frame; d) Mobile communication module, used for uplink and downlink communication with the cloud computing module; The cloud computing module is used to perform feature point detection and matching algorithms on the first frame images captured by the left and right cameras and send the initial coordinates of the tracking target; based on the sub-pixel precision coordinates of the tracking target in the nth frame left and right images and the left and right projection matrices, it performs triangulation to solve the three-dimensional coordinates of the tracking target and outputs the results; The left camera subsystem and the right camera subsystem do not require a wired connection and do not need to communicate with each other. The local image processing and trigger control module communicates with the camera module, GNSS timing module, and mobile communication module via local wired connections. The external trigger of the camera module is generated by the local image processing and trigger control module based on the GNSS pulse-per-second (PPS) signal.

Images