Image virtual horizontal calibration method and system for power monitoring control ball

Abstract

This invention proposes a virtual horizontal calibration method and system for power monitoring surveillance cameras, relating to the technical field of image geometric correction. The method involves acquiring real-time acceleration data and calculating the tilt angle. The variance of the tilt angle within a sliding window is calculated. If the variance exceeds a certain threshold, the previously cached image geometric correction matrix is ​​forcibly reused. Otherwise, a dual verification is performed using a tilt angle change micro-motion threshold and the maximum effective cache duration. The image geometric correction matrix is ​​reconstructed only when the tilt angle change exceeds the micro-motion threshold or the cache duration exceeds the maximum effective duration; otherwise, the cached image geometric correction matrix continues to be reused. Finally, the currently acquired video image from the power monitoring surveillance camera is transformed to obtain a horizontally calibrated virtual horizontal viewpoint image. This solution significantly reduces the computational power consumption of the embedded processor while eliminating the cumulative errors generated by the long-term operation of the power monitoring surveillance camera, thus improving the quality of the final acquired image.

Description

Technical Field

[0001] This invention relates to the technical field of image geometric correction, and more specifically, to a method and system for virtual horizontal calibration of images for power monitoring surveillance cameras. Background Technology

[0002] A power monitoring PTZ camera is an intelligent mobile monitoring device specifically designed for the power industry. It integrates video image acquisition and wireless transmission functions, and is used in scenarios such as power line inspection, emergency repair, and construction safety supervision. Its core features are rapid deployment, strong environmental adaptability, and a high degree of intelligence, addressing the pain points of complex environments and the difficulty of fixed monitoring coverage in power operations.

[0003] In practical applications, the deployment of conventional portable power monitoring and control cameras and other intelligent vision devices typically relies on operators manually adjusting mechanical support structures such as tripods to ensure the equipment is level. However, in complex environments requiring rapid deployment and with uneven terrain, such as power emergency repairs and field construction, operators find it difficult to achieve ideal level placement of the equipment manually. Even if the initial installation barely meets the specifications, the intelligent vision device is easily moved during subsequent operations due to unintentional collisions by on-site personnel, cable dragging, or temporary movement that obstructs construction, causing its posture to change and fall back into a tilted state. In such cases, mechanical leveling of the power monitoring and control camera is required, but this leveling process has a slow response time and is difficult to achieve dynamic adjustment.

[0004] Once the surveillance camera is not in a horizontal position, the video images it captures will inevitably suffer from geometric distortion and perspective errors. This uncorrected tilted posture destroys the geometric features of the image, seriously interfering with subsequent intelligent video analysis based on deep learning. In particular, for behavior analysis scenarios that rely on verticality features, the tilted image will severely distort the vertical relationship of objects relative to the ground reference frame, which can easily lead to state recognition errors. For example, it may mistakenly identify construction workers who are standing normally on the ground as climbing or working at height, thus causing frequent false alarms and seriously reducing the practicality and reliability of the intelligent monitoring system in actual complex working conditions.

[0005] Therefore, ensuring that the images captured by the surveillance camera are straight and avoiding image distortion caused by device tilt is a technical problem that urgently needs to be solved by current power monitoring surveillance cameras. Summary of the Invention

[0006] To address the issues of high computational power consumption and low image acquisition quality caused by the tilt of power monitoring surveillance cameras, this invention proposes a virtual horizontal calibration method and system for power monitoring surveillance cameras. It considers how to reuse the image geometric correction matrix to ensure the correct viewing angle of the acquired images, avoiding image distortion caused by the tilt of the surveillance camera, reducing computational power consumption while improving the quality of the final acquired images.

[0007] To achieve the above-mentioned technical effects, the technical solution of the present invention is as follows:

[0008] Firstly, this application proposes a virtual horizontal calibration method for images of power monitoring surveillance cameras, comprising the following steps:

[0009] S1. Collect real-time acceleration data in the three axes of the current power monitoring deployment ball, and preprocess the collected acceleration data;

[0010] S2. Based on the preprocessed acceleration data, calculate the tilt angle of the power monitoring and control ball at the current moment;

[0011] S3. Based on the tilt angle of the power monitoring satellite at the current moment, construct and cache the image geometric correction matrix of the power monitoring satellite at the current moment, and execute S4;

[0012] S4. Calculate the variance of the tilt angle within the preset frame sliding window at the current time, and determine whether the variance of the tilt angle is less than the preset stable state safety threshold. If yes, the power monitoring deployment ball is in a stable state, and proceed to S6; otherwise, the power monitoring deployment ball is in an unstable state, and proceed to step S5.

[0013] S5. Use the image geometric correction matrix cached at the previous time step as the image geometric correction matrix at the current time step, and execute S7;

[0014] S6. Calculate the difference between the tilt angle at the current moment and the tilt angle at the previous moment. If the difference is less than the preset micro-motion threshold, use the image geometric correction matrix cached at the previous moment as the image geometric correction matrix at the current moment and execute S7. Otherwise, return to S3.

[0015] S7. Determine whether the cache duration of the image geometric correction matrix exceeds the maximum valid duration. If yes, reconstruct the image geometric correction matrix of the current pose and execute S8; otherwise, use the cached image geometric correction matrix of the previous time step as the image geometric correction matrix of the current time step and execute S8.

[0016] S8. Based on the image geometric correction matrix, transform the currently acquired video image from the power monitoring satellite to obtain a virtual horizontal view image after horizontal calibration.

[0017] In this technical solution, real-time acceleration data of the power monitoring sphere is first acquired and the tilt angle is calculated. Then, the variance of the tilt angle within a sliding window is calculated to evaluate the stability of the power monitoring sphere. If the variance exceeds a preset standard, it is determined to be in an unstable state, and the geometric correction matrix of the previous frame image is forcibly reused. If it is in a stable state, a dual verification is performed using a tilt angle change micro-motion threshold and the maximum effective buffer duration. The image geometric correction matrix is ​​reconstructed only when the tilt angle change exceeds the micro-motion threshold or the buffer duration exceeds the maximum effective duration. Otherwise, the buffered image geometric correction matrix continues to be reused, and the currently acquired video image from the power monitoring sphere is transformed to obtain a horizontally calibrated virtual horizontal viewpoint image. This solution significantly reduces the computing power consumption of the embedded processor while eliminating the cumulative errors generated by the long-term operation of the power monitoring sphere, thus improving the quality of the final acquired image.

[0018] Preferably, the system collects real-time acceleration data along three axes from the current power monitoring and control sphere, and preprocesses the collected acceleration data, including:

[0019] Collect raw acceleration data in the three axes of the current power monitoring and control sphere. ;

[0020] Using a pre-calibrated alignment matrix The raw acceleration data is transformed to calculate the acceleration vector in the coordinate system of the power monitoring PTZ camera. The expression is:

[0021]

[0022] in, The components in the three-axis directions are represented as follows: , express Acceleration data in the axial direction, express Acceleration data in the axial direction, express Acceleration data in the axial direction.

[0023] Preferably, the tilt angle includes a pitch angle and a roll angle; calculating the tilt angle of the current attitude of the power monitoring and control ball includes: calculating the pitch angle and roll angle of the current attitude of the power monitoring and control ball in real time, with the following expressions:

[0024]

[0025]

[0026] in, This represents the pitch angle at a specific moment in the original video image from the power monitoring PTZ camera. This represents the roll angle at a specific moment in the original video image of the power monitoring and control sphere.

[0027] Preferably, the step of constructing and caching the image geometric correction matrix based on the tilt angle of the power monitoring sphere at the current moment includes:

[0028] Construct a tilt-compensated rotation submatrix that tilts in the opposite direction of the roll angle. And the tilt-compensating rotator matrix tilted in the opposite direction of the pitch angle. The expression is:

[0029]

[0030]

[0031] Based on tilt compensation rotation submatrix and tilt-compensated rotation submatrix Construct a tilt-compensated rotation matrix The expression is:

[0032] ;

[0033] Let the horizontal resolution of the video images acquired by the power monitoring PTZ camera be . and image vertical resolution Based on horizontal resolution and image vertical resolution Obtain the camera parameter matrix of the power monitoring PTZ camera. The expression is:

[0034]

[0035] in, , , Construct the image geometric correction matrix The expression is:

[0036] .

[0037] Preferably, calculating the variance of the tilt angle within the preset frame sliding window at the current time includes:

[0038] Calculate the frame interval of the sliding window using moving average filtering. The average pitch angle and average roll angle are expressed as follows:

[0039]

[0040]

[0041] in, This indicates a pre-set smooth window;

[0042] Based on the average pitch angle and average roll angle, the variance of the pitch angle of the current attitude falling within the preset sliding window is calculated. variance of roll angle The expressions are as follows:

[0043]

[0044]

[0045] in, This represents the deviation between the pitch angle and the average pitch angle at each moment under the current image geometric correction matrix. This indicates the deviation between the roll angle and the average roll angle at each time step under the current image geometric correction matrix;

[0046] The variance of the tilt angle is taken as the variance of the pitch angle. variance of roll angle The maximum value in.

[0047] Preferably, the difference between the tilt angle of the current attitude and the tilt angle of the previous attitude is calculated. The expression is:

[0048]

[0049] in, The pitch angle represents the attitude at the current moment. The pitch angle represents the attitude at the previous moment. The roll angle represents the attitude at the current moment. This indicates the roll angle of the attitude at the previous moment.

[0050] Preferably, when using the image geometric correction matrix cached at the previous time step as the image geometric correction matrix at the current time step, the method further includes using a matrix difference algorithm to smoothly transition the image geometric correction matrix from the previous time step to the image geometric correction matrix cached at the current time step, including:

[0051] Preset the number of frames required for matrix transition The interpolation coefficients are calculated using the following expression:

[0052]

[0053] in, Indicates the current frame index during the matrix smooth transition process;

[0054] Based on the interpolation coefficients, the current image geometric correction matrix, and the cached image geometric correction matrix from the previous time step, the image geometric correction matrix for the k-th frame during matrix transition is calculated. The expression is:

[0055]

[0056] in, This represents the geometric correction matrix of the image at the previous time step. This represents the image geometric correction matrix cached at the current time.

[0057] Preferably, the image virtual horizontal calibration method for power monitoring surveillance spheres further includes: performing vertical climbing behavior analysis on the monitoring targets of the power monitoring surveillance sphere based on the horizontally calibrated virtual horizontal view image, including:

[0058] Perform target detection on the horizontally calibrated virtual horizontal viewpoint image to obtain the target bounding box;

[0059] Extract the ordinate of the bottom edge of the target bounding box Based on the preset pixel-to-length conversion factor And the virtual horizontal view image after horizontal calibration Calculate the vertical coordinate as the actual ground altitude of the target. The expression is:

[0060]

[0061] Determine whether the actual height above the ground exceeds a preset height threshold; if so, the monitored target is exhibiting climbing behavior.

[0062] If not, then determine the actual height above the ground. Whether the height exceeds 0.8 times the preset height threshold and whether the vertical speed of the monitored target exceeds the preset minimum speed; if so, the monitored target is exhibiting climbing behavior.

[0063] In other cases, the monitored target does not engage in any climbing behavior.

[0064] Secondly, this application also proposes an image virtual horizontal calibration system for power monitoring surveillance cameras. The system is used to implement an image virtual horizontal calibration method for power monitoring surveillance cameras, and the system includes:

[0065] The data acquisition and preprocessing unit is used to acquire real-time acceleration data in the three axes of the current power monitoring and control sphere, and to preprocess the acquired acceleration data.

[0066] The tilt angle calculation unit is used to calculate the tilt angle of the current attitude of the power monitoring deployment ball based on the preprocessed acceleration data.

[0067] The image geometric correction matrix construction unit is used to construct and cache the image geometric correction matrix of the power monitoring deployment ball at the current moment based on the tilt angle of the power monitoring deployment ball at the current moment.

[0068] The adaptive cache control unit is used to calculate the variance of the tilt angle within the sliding window of the preset attitude frame at the current moment. Based on the comparison between the variance of the tilt angle and the preset stable state safety threshold, it determines whether the power monitoring deployment ball is in a stable state. If the power monitoring deployment ball is in an unstable state, the image geometric correction matrix cached at the previous moment is used as the image geometric correction matrix at the current moment. If the power monitoring deployment ball is in a stable state, the difference between the tilt angle of the attitude at the current moment and the tilt angle of the attitude at the previous moment is calculated. Based on the comparison between the difference and the preset micro-motion threshold, it determines whether the image set correction matrix at the current moment needs to be updated.

[0069] The cache duration judgment unit is used to determine whether the cache duration of the image geometric correction matrix exceeds the maximum effective duration. If it exceeds the maximum effective duration, the image geometric correction matrix of the pose at the current time is reconstructed; if it does not exceed the maximum effective duration, the image geometric correction matrix cached at the previous time is used as the image geometric correction matrix at the current time.

[0070] The horizontal calibration unit is used to transform the currently acquired video image from the power monitoring and control sphere based on the image geometric correction matrix to obtain a virtual horizontal view image after horizontal calibration.

[0071] Thirdly, this application proposes an image virtual horizontal calibration device for power monitoring surveillance cameras, the device being used to implement an image virtual horizontal calibration method for power monitoring surveillance cameras, comprising:

[0072] The data acquisition and preprocessing module is used to acquire real-time acceleration data in the three axes of the current power monitoring deployment ball, and to preprocess the acquired acceleration data.

[0073] The tilt angle calculation module is used to calculate the tilt angle of the current attitude of the power monitoring deployment ball based on the preprocessed acceleration data.

[0074] The image geometric correction matrix construction module is used to construct and cache the image geometric correction matrix of the power monitoring deployment ball at the current moment based on the tilt angle of the power monitoring deployment ball at the current moment.

[0075] The adaptive cache control module is used to calculate the variance of the tilt angle within the sliding window of the preset attitude frame at the current moment. Based on the comparison between the variance of the tilt angle and the preset stable state safety threshold, it determines whether the power monitoring deployment ball is in a stable state. If the power monitoring deployment ball is in an unstable state, the image geometric correction matrix cached at the previous moment is used as the image geometric correction matrix at the current moment. If the power monitoring deployment ball is in a stable state, the difference between the tilt angle of the attitude at the current moment and the tilt angle of the attitude at the previous moment is calculated. Based on the comparison between the difference and the preset micro-motion threshold, it determines whether the image set correction matrix at the current moment needs to be updated.

[0076] The cache duration judgment module is used to determine whether the cache duration of the image geometric correction matrix exceeds the maximum effective duration. If it exceeds the maximum effective duration, the image geometric correction matrix of the current pose is reconstructed; if it does not exceed the maximum effective duration, the image geometric correction matrix cached at the previous time step is used as the image geometric correction matrix of the current time step.

[0077] The horizontal calibration module is used to transform the currently acquired video image from the power monitoring sphere based on the image geometric correction matrix to obtain a virtual horizontal view image after horizontal calibration.

[0078] Compared with the prior art, the beneficial effects of the present invention are:

[0079] This invention proposes a virtual horizontal calibration method and system for power monitoring surveillance cameras, relating to the technical field of image geometric correction. First, real-time acceleration data is acquired and the tilt angle is calculated. Then, the variance of the tilt angle within a sliding window is calculated to assess device stability. If the variance exceeds the standard, the system is deemed unstable, and the geometric correction matrix of the previous frame is forcibly reused. If the system is stable, a dual verification is performed using a tilt angle change micro-motion threshold and the maximum effective buffer duration. The image geometric correction matrix is ​​reconstructed only when the tilt angle change exceeds the micro-motion threshold or the buffer duration exceeds the maximum effective duration; otherwise, the buffered image geometric correction matrix continues to be reused. The currently acquired video image from the power monitoring surveillance camera is then transformed to obtain a horizontally calibrated virtual horizontal viewpoint image. This solution significantly reduces the computational power consumption of the embedded processor while eliminating the cumulative errors generated by the long-term operation of the power monitoring surveillance camera, thus improving the quality of the final acquired image. Attached Figure Description

[0080] Figure 1 This is a flowchart illustrating the image virtual horizontal calibration method for power monitoring surveillance cameras proposed in Embodiment 1 of the present invention;

[0081] Figure 2 This is a schematic diagram illustrating the process of performing height-climbing behavior analysis on the monitoring target of the power monitoring deployment ball based on the virtual horizontal view image after horizontal calibration, as proposed in Embodiment 2 of the present invention.

[0082] Figure 3 This is a schematic diagram of the image virtual horizontal calibration system for power monitoring and control spheres proposed in Embodiment 3 of the present invention;

[0083] Figure 4 This is a schematic diagram of the image virtual horizontal calibration device for power monitoring and control spheres proposed in Embodiment 3 of the present invention. Detailed Implementation

[0084] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.

[0085] To better illustrate this embodiment, some parts of the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions;

[0086] It is understandable to those skilled in the art that some well-known details may be omitted from the accompanying drawings.

[0087] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0088] The positional relationships depicted in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0089] Example 1

[0090] This embodiment proposes a virtual horizontal calibration method for power monitoring PTZ cameras. A flowchart illustrating this method can be found here. Figure 1 This includes the following steps:

[0091] S1. Collect real-time acceleration data in the three axes of the current power monitoring deployment ball, and preprocess the collected acceleration data;

[0092] S2. Based on the preprocessed acceleration data, calculate the tilt angle of the power monitoring and control ball at the current moment;

[0093] S3. Based on the tilt angle of the power monitoring satellite at the current moment, construct and cache the image geometric correction matrix of the power monitoring satellite at the current moment, and execute S4;

[0094] S4. Calculate the variance of the tilt angle within the preset frame sliding window at the current time, and determine whether the variance of the tilt angle is less than the preset stable state safety threshold. If yes, the power monitoring deployment ball is in a stable state, and proceed to S6; otherwise, the power monitoring deployment ball is in an unstable state, and proceed to step S5.

[0095] S5. Use the image geometric correction matrix cached at the previous time step as the image geometric correction matrix at the current time step, and execute S7;

[0096] S6. Calculate the difference between the tilt angle at the current moment and the tilt angle at the previous moment. If the difference is less than the preset micro-motion threshold, use the image geometric correction matrix cached at the previous moment as the image geometric correction matrix at the current moment and execute S7. Otherwise, return to S3.

[0097] S7. Determine whether the cache duration of the image geometric correction matrix exceeds the maximum valid duration. If yes, reconstruct the image geometric correction matrix of the current pose and execute S8; otherwise, use the cached image geometric correction matrix of the previous time step as the image geometric correction matrix of the current time step and execute S8.

[0098] S8. Based on the image geometric correction matrix, transform the currently acquired video image from the power monitoring satellite to obtain a virtual horizontal view image after horizontal calibration.

[0099] Specifically, if only angle difference is relied upon for determination, matrix updates may not be triggered for an extended period when the device experiences extremely slow attitude drift. This causes the displayed "horizontal state" to gradually deviate from the actual physical state, affecting the accuracy of subsequent behavior analysis. Therefore, a maximum effective duration is set for the cache matrix. The typical value for the maximum effective duration is 3-10 seconds; exceeding this duration will force a matrix update. The current system time is recorded each time the cache is updated using a new image geometric correction matrix. During each frame processing, based on the current time and the recorded system time Calculate the duration of cache usage. ;when ≥ Even at that time < It still forces the matrix to be rebuilt and the cache to be updated.

[0100] In this embodiment, real-time acceleration data is first acquired and the tilt angle is calculated. Then, the tilt angle variance within a sliding window is calculated to assess device stability. If the variance exceeds the standard, the system is deemed unstable, and the geometric correction matrix of the previous frame is forcibly reused. If the system is stable, a dual verification is performed using a tilt angle change micro-motion threshold and the maximum effective buffer duration. The image geometric correction matrix is ​​reconstructed only when the tilt angle change exceeds the micro-motion threshold or the buffer duration exceeds the maximum effective duration; otherwise, the buffered image geometric correction matrix continues to be reused. The currently acquired video image from the power monitoring sphere is then transformed to obtain a horizontally calibrated virtual horizontal viewpoint image. This solution significantly reduces the computational power consumption of the embedded processor while eliminating the cumulative errors generated by the long-term operation of the power monitoring sphere, thus improving the quality of the final acquired image.

[0101] Example 2

[0102] In this embodiment, real-time acceleration data along the three axes of the current power monitoring and control sphere is collected, and the collected acceleration data is preprocessed, including:

[0103] Collect raw acceleration data in the three axes of the current power monitoring and control sphere. ;

[0104] Using a pre-calibrated alignment matrix The raw acceleration data is transformed to calculate the acceleration vector in the coordinate system of the power monitoring PTZ camera. The expression is:

[0105]

[0106] in, The components in the three-axis directions are represented as follows: , express Acceleration data in the axial direction, express Acceleration data in the axial direction, express Acceleration data in the axial direction.

[0107] Specifically, the processor of the power monitoring surveillance sphere periodically acquires the raw register values ​​of the triaxial accelerometer in the inertial measurement unit (IMU) inside the sphere at a preset sampling frequency via an onboard high-speed digital communication interface. Subsequently, the processor linearly converts the acquired raw register values ​​into physically meaningful raw acceleration data based on the sensitivity coefficients specified in the sensor specifications. Obtain the raw acceleration data at the current moment. ;

[0108] Specifically, since there may be minor vibrations at power construction sites caused by wind or passing vehicles, directly using the raw acceleration value would result in frequent jumps in the calculated tilt angle, leading to "high-frequency jitter" in the video image. Therefore, this embodiment uses a moving average filter algorithm for noise reduction.

[0109] Specifically, let the acceleration vector in the coordinate system of the power monitoring PTZ camera be defined. The components in the three-axis directions are represented as follows: Its unit is usually normalized to gravitational acceleration. or physical units In terms of physics, the specific force measured by a triaxial accelerometer is the difference between the gravitational acceleration vector and the acceleration vector of the moving equipment.

[0110] Under static or quasi-static conditions, since the control ball is mounted on a relatively fixed platform (such as a tripod or a stationary vehicle), its linear acceleration relative to gravitational acceleration... This is extremely small and can be approximated as negligible. At this point, the vector output by the sensor... In physical essence, it is equivalent to the gravitational acceleration vector. The inverse projection in the device's own three-dimensional coordinate system. In other words, This set of values ​​accurately records the direction of gravity relative to the device. Geometric components of each axis, Mainly caused by gravitational acceleration It is obtained by projection after rotation in the equipment coordinate system.

[0111] In this embodiment, the tilt angle includes pitch angle and roll angle; calculating the tilt angle of the current attitude of the power monitoring and control ball includes: calculating the pitch angle and roll angle of the current attitude of the power monitoring and control ball in real time, with the following expressions:

[0112]

[0113]

[0114] in, This represents the pitch angle at a specific moment in the original video image from the power monitoring PTZ camera. This represents the roll angle at a specific moment in the original video image of the power monitoring and control sphere.

[0115] Specifically, the pitch angle The power monitoring and control sphere surrounds itself The rotation angle of the shaft; the roll angle The power monitoring and control sphere surrounds itself The rotation angle of the axis.

[0116] In this embodiment, the step of constructing and caching the image geometric correction matrix based on the tilt angle of the power monitoring sphere at the current moment includes:

[0117] Construct a tilt-compensated rotation submatrix that tilts in the opposite direction of the roll angle. And the tilt-compensating rotator matrix tilted in the opposite direction of the pitch angle. The expression is:

[0118]

[0119]

[0120] Based on tilt compensation rotation submatrix and tilt-compensated rotation submatrix Construct a tilt-compensated rotation matrix The expression is:

[0121] ;

[0122] Let the horizontal resolution of the video images acquired by the power monitoring PTZ camera be . and image vertical resolution Based on horizontal resolution and image vertical resolution Obtain the camera parameter matrix of the power monitoring PTZ camera. The expression is:

[0123]

[0124] in, , , Construct the image geometric correction matrix The expression is:

[0125] .

[0126] Specifically, the tilt compensation rotation matrix is ​​a The linear transformation matrix, in its physical essence, is a mathematical description of a virtual reverse rotation of the camera's imaging plane. Its core function is to simulate a rotational motion completely opposite to the physical tilt direction through algorithms, without changing the camera's physical orientation, thereby counteracting the effects of gravity on the image at the digital image level.

[0127] The pitch angle calculated based on the aforementioned steps and roll angle This reflects the current state of the equipment deviating from the ideal horizontal plane. To "straighten" the image, a rotation of equal magnitude but opposite direction must be applied. If the equipment is physically rotating... The axis pitched downwards For example, if the head nods at a 30-degree angle, then the compensation matrix must allow the image to rotate... Axial upward rotation (That is, tilting the head up 30 degrees). If the equipment is physically rotated... The axis tilted to the right. (For example, if the image is tilted 10 degrees to the right), then the compensation matrix must make the image rotate around... Axial rotation to the left (i.e., 10 degrees to the left).

[0128] Since the tilting of equipment is usually a compound event (both nodding and tilting), the tilt compensation rotation matrix It is the product of the two sub-matrices mentioned above. According to the order of matrix multiplication (usually following ZYX or similar Euler angle rules; in this embodiment, it is preferred to compensate for roll first and then pitch, or according to the actual defined coordinate system order), the final tilt compensation rotation matrix is... The calculation expression is:

[0129]

[0130] The tilt compensation rotation matrix Ultimately, it contains 9 floating-point elements. In subsequent image processing, it will be "embedded" into the image geometric correction matrix. The intermediate layer. When the pixel coordinate vector of the original image is processed by this matrix, the pixels will be remapped in space. The effect is equivalent to forcibly "twisting" the optical axis of the camera to the horizontal direction, so that the direction of gravity (Z axis) in the output image is strictly perpendicular to the horizon, realizing the effect of "software-defined virtual gimbal".

[0131] Specifically, constructing the image geometric correction matrix This is a crucial step in achieving virtual horizontal calibration. This step aims to establish a mathematical model that accurately maps pixels in the original tilted image coordinate system to their corresponding positions in the virtual horizontal image coordinate system. First, the system needs to establish a pinhole imaging model of the power monitoring PTZ camera. This model is achieved through the camera intrinsic parameter matrix. This describes the geometric relationship of a point in three-dimensional space projected onto a two-dimensional image plane. It is typically assumed that the optical center of the camera is located at the geometric center of the image plane. Therefore, based on the horizontal resolution of the acquired raw video image... and vertical resolution The coordinates of the optical center can be calculated directly:

[0132] Specifically, in engineering scenarios lacking expensive and precise calibration procedures, this invention employs a general estimation strategy based on the field of view. The diagonal length of the image is used as a reference benchmark for the equivalent focal length, i.e., assuming... This setting implies that the camera's field of view is within a certain standard range, which is sufficient to meet the geometric correction accuracy requirements of most security monitoring scenarios, avoiding the tedious checkerboard calibration process.

[0133] Specifically, the image geometric correction matrix The synthesis logic follows a closed-loop mathematical transformation principle: "inverse mapping from pixel coordinate space to 3D physical space, performing pose compensation rotation in 3D space, and then remapping back to pixel coordinate space." Specifically, this matrix is ​​essentially an operator describing the homography transformation relationship between the original tilted imaging plane and the target virtual horizontal imaging plane. Its construction process consists of three consecutive linear transformation steps: First, using the inverse matrix of the camera intrinsic parameter matrix... Inverse projection is performed on the pixel coordinates of the original image to restore the pixels on the two-dimensional plane to normalized three-dimensional ray vectors in the camera coordinate system, thereby recovering the physical geometric properties of the scene; then, a pre-constructed tilt compensation rotation matrix is ​​used... A reverse rotation operation is applied to the recovered 3D ray vectors, mathematically simulating forcibly twisting the camera's optical axis from its actual tilt state to an ideal horizontal state, thus offsetting the pitch and roll deviations in the physical world; finally, the camera intrinsic parameter matrix is ​​used... A reprojection operation is performed on the rotated and corrected 3D ray vectors to remap them back to the 2D pixel coordinate system, thus determining the corrected pixel positions. According to the associative law of matrix multiplication, the matrices corresponding to the above three steps are multiplied sequentially, i.e. This synthesizes the final image geometric correction matrix, which can complete the perspective transformation correction of all pixels in the image at one time with extremely high computational efficiency, realizing the mathematical reconstruction from the tilted viewpoint to the standard horizontal viewpoint.

[0134] In this embodiment, calculating the variance of the tilt angle within the preset frame sliding window at the current time includes:

[0135] Calculate the frame interval of the sliding window using moving average filtering. The average pitch angle and average roll angle are expressed as follows:

[0136]

[0137]

[0138] in, This indicates a pre-set smooth window;

[0139] Based on the average pitch angle and average roll angle, the variance of the pitch angle of the current attitude falling within the preset sliding window is calculated. variance of roll angle The expressions are as follows:

[0140]

[0141]

[0142] in, This represents the deviation between the pitch angle and the average pitch angle at each moment under the current image geometric correction matrix. This indicates the deviation between the roll angle and the average roll angle at each time step under the current image geometric correction matrix;

[0143] The variance of the tilt angle is taken as the variance of the pitch angle. variance of roll angle The maximum value in.

[0144] Specifically, the preset stable state safety threshold should be set in conjunction with the outdoor power application scenario, with a typical stable state safety threshold of 0.0005~0.002.

[0145] If the equipment is determined to have been subjected to severe vibration or impact, the following handling strategy should be adopted:

[0146] First, pause matrix updates and maintain the correction matrix from the previous stable time step. The process remains unchanged and continues to be used for image correction in the current frame; data acquisition continues: the sliding window data is still updated, and variance changes are continuously monitored; when the variance falls below the threshold, the normal update judgment process is automatically restored.

[0147] Specifically, in practical applications, the accelerometers mounted on power monitoring surveillance spheres are highly sensitive to high-frequency vibrations. Even minor breezes at construction sites, ground vibrations caused by passing vehicles, or thermal noise from the electronic components themselves can affect the calculated pitch and roll angles. and Slight changes occur within a short period. If the original angle is used directly for image correction, the video image will exhibit high-frequency jitter, causing dizziness for the observer. Therefore, this embodiment introduces a moving average filtering mechanism, which acts as a low-voltage filter to remove high-frequency noise and retain low-frequency, true attitude changes.

[0148] Specifically, smooth window The value of has a critical impact on system performance, and this embodiment provides... Basis for value selection:

[0149] If smooth window If the value is too small, the noise suppression effect will be insignificant, and residual jitter may still remain in the image. However, the system will respond quickly to the actual movement of the device with low lag. If the smooth window... If the value is too large, the image will be very smooth and stable, but it will produce noticeable lag. When the device's posture changes significantly, the image will require a longer adjustment time, resulting in high lag.

[0150] Specifically, to further enhance the system's robustness in complex dynamic environments, this embodiment introduces an adaptive confidence assessment mechanism based on analysis of variance. Specifically, the system calculates the statistical variance of historical pitch and roll angle data within a sliding window in real time, using this as a quantitative indicator of the current physical stability of the device's attitude. Subsequently, a mapping relationship is established using an inverse proportional function or a Gaussian decay model to convert the calculated angle variance into a numerical range within a certain range. Normalized confidence coefficient of the interval This establishes a negative correlation logic: "the smaller the variance, the more reliable the data; the larger the variance, the more questionable the data." In subsequent anomaly handling steps, this confidence coefficient... It acts as a "gating switch" for updating the image geometric correction model: when the system detects a sudden vibration of the equipment caused by strong winds, accidental impacts, or passing vehicles, resulting in a sharp increase in the angle variance and thus affecting the confidence level... When the data falls below the preset safety threshold, the system will immediately trigger the abnormal protection strategy, determine that the sensor data at the current moment is invalid noise, and forcibly freeze the update of the perspective transformation matrix. It will directly reuse the correction model from the previous high confidence moment until the physical vibration subsides and the confidence level returns to normal. This effectively prevents the monitoring screen from shaking violently or tearing due to sudden physical disturbances, and ensures the smoothness and continuity of the virtual horizontal perspective image output.

[0151] In this embodiment, the difference between the tilt angle of the current attitude and the tilt angle of the previous attitude is calculated. The expression is:

[0152]

[0153] in, The pitch angle represents the attitude at the current moment. The pitch angle represents the attitude at the previous moment. The roll angle represents the attitude at the current moment. This indicates the roll angle of the attitude at the previous moment.

[0154] Specifically, using Euclidean distance can comprehensively consider changes in both pitch and roll directions. This method constructs a two-dimensional angle vector space and calculates the vector magnitude as the total attitude deviation, which can effectively capture the composite tilt component of the device in any direction. This avoids the sensitivity loss problem of single-dimensional threshold judgment in the diagonal direction, thus ensuring the isotropic and robustness of the virtual horizontal correction triggering mechanism.

[0155] Specifically, the typical value of the preset tilt angle micro-motion threshold is 0.005 ~ 0.015 radians.

[0156] Specifically, to significantly reduce the computational load on the embedded processor and improve the visual stability of the video while ensuring correction accuracy, the system introduces an image geometric correction matrix cache reuse mechanism based on an angle change threshold. Specifically, this mechanism is based on the physical fact that "the video frame rate is much higher than the frequency of device attitude changes." The system allocates dedicated storage space in memory to store the image geometric correction matrix and its corresponding reference tilt angle that were constructed in the previous moment. Within each frame processing cycle, the processor first calculates the composite difference between the tilt angle (including pitch and roll angles) calculated at the current moment and the reference tilt angle cached at the previous moment, and compares this composite difference with a preset tilt angle change threshold. This threshold sets a "dead zone" for attitude changes, used to filter out non-essential angle fluctuations caused by sensor thermal noise or environmental micro-vibrations (such as high-frequency micro-flickers caused by wind). If the calculated angle difference is less than the threshold, the system determines that the device is currently in a relatively stationary or steady state. In this case, it directly calls the image geometric correction matrix cached in memory from the previous moment as the transformation operator for the current frame, skipping the complex matrix construction process. This "zero-order hold" strategy not only eliminates redundant calculations but also fundamentally avoids the "breathing effect" or pixel jitter at the edges of the image caused by overly sensitive correction. Conversely, if the difference exceeds the threshold, the system determines that the device has undergone a substantial change in attitude. It then immediately reconstructs the image geometric correction matrix based on the latest tilt angle, corrects the video, and synchronously updates the cached matrix and reference angle to ensure that the monitoring screen can respond to the actual movement of the device in real time.

[0157] In this embodiment, when using the image geometric correction matrix cached at the previous time step as the image geometric correction matrix at the current time step, the method further includes using a matrix difference algorithm to smoothly transition the image geometric correction matrix from the previous time step to the image geometric correction matrix cached at the current time step, including:

[0158] Preset the number of frames required for matrix transition The interpolation coefficients are calculated using the following expression:

[0159]

[0160] in, Indicates the current frame index during the matrix smooth transition process;

[0161] Based on the interpolation coefficients, the current image geometric correction matrix, and the cached image geometric correction matrix from the previous time step, the image geometric correction matrix for the k-th frame during matrix transition is calculated. The expression is:

[0162]

[0163] in, This represents the geometric correction matrix of the image at the previous time step. This represents the image geometric correction matrix cached at the current time.

[0164] Specifically, specific interpolation algorithms include, but are not limited to: matrix linear interpolation, spherical linear interpolation, and easing interpolation based on nonlinear functions.

[0165] The technical effect of this step is to eliminate video screen tearing or jumps caused by sudden changes in matrix parameters, maintain the temporal continuity of the video stream, and thus ensure the target tracking stability of downstream intelligent analysis algorithms.

[0166] In this embodiment, the method further includes: performing specific behavioral analysis on the monitoring target of the power monitoring surveillance sphere based on the horizontally calibrated virtual horizontal view image. Taking the climbing behavior analysis as an example, the flowchart of the climbing behavior analysis is shown below. Figure 2 This includes the following steps:

[0167] Perform target detection on the horizontally calibrated virtual horizontal viewpoint image to obtain the target bounding box;

[0168] Extract the ordinate of the bottom edge of the target bounding box Based on the preset pixel-to-length conversion factor And the virtual horizontal view image after horizontal calibration Calculate the vertical coordinate as the actual ground altitude of the target. The expression is:

[0169]

[0170] Determine whether the actual height above the ground exceeds a preset height threshold; if so, the monitored target is exhibiting climbing behavior.

[0171] If not, then determine the actual height above the ground. Whether the height exceeds 0.8 times the preset height threshold and whether the vertical speed of the monitored target exceeds the preset minimum speed; if so, the monitored target is exhibiting climbing behavior.

[0172] In other cases, the monitored target does not engage in any climbing behavior.

[0173] Specifically, the expression for determining the height is:

[0174]

[0175] in, This represents a binary parameter used to determine whether climbing behavior has occurred. This indicates a preset height threshold. Indicates the preset minimum speed;

[0176] The logic for determining altitude employs a dual-verification strategy combining a "static position threshold" and "dynamic trend compensation" to address the issues of missed detections or lags that can easily occur in critical regions with a single altitude determination. This logic is configured in the processor to execute two decision branches in parallel: first, it performs an absolute altitude determination; if the detected current target relative altitude... It directly exceeded the preset absolute safety height threshold. Regardless of the target's movement state, the system classifies it as a definite ascending behavior (True) to ensure zero-tolerance capture of obvious violations. Secondly, the system introduces dynamic compensation judgment for critical warning zones; if the current height... Although it has not completely broken through the absolute threshold, it has entered the critical high-risk zone (specifically reflected in the current relative height of the target). Greater than 0.8 times the threshold, that is Furthermore, a significant upward vertical movement trend (i.e., vertical velocity) was detected in the target. Greater than the minimum effective speed threshold The system will predict the movement intention and preemptively determine it as climbing behavior (True). This logical design can effectively cover dynamic scenarios of "rapid climbing but not yet reaching the top", realizing the transformation from passive monitoring to active early warning. In other cases where neither of the above two conditions is met, the system outputs No (False) and determines that it is not a safe climbing state.

[0177] Example 3

[0178] This embodiment proposes a virtual horizontal calibration system for images from power monitoring surveillance cameras. In this embodiment, the system is used to implement a virtual horizontal calibration method for images from power monitoring surveillance cameras. A schematic diagram of the system is shown below. Figure 3 As shown, it includes:

[0179] The data acquisition and preprocessing unit is used to acquire real-time acceleration data in the three axes of the current power monitoring and control sphere, and to preprocess the acquired acceleration data.

[0180] The tilt angle calculation unit is used to calculate the tilt angle of the current attitude of the power monitoring deployment ball based on the preprocessed acceleration data.

[0181] The image geometric correction matrix construction unit is used to construct and cache the image geometric correction matrix of the power monitoring deployment ball at the current moment based on the tilt angle of the power monitoring deployment ball at the current moment.

[0182] The adaptive cache control unit is used to calculate the variance of the tilt angle within the sliding window of the preset attitude frame at the current moment. Based on the comparison between the variance of the tilt angle and the preset stable state safety threshold, it determines whether the power monitoring deployment ball is in a stable state. If the power monitoring deployment ball is in an unstable state, the image geometric correction matrix cached at the previous moment is used as the image geometric correction matrix at the current moment. If the power monitoring deployment ball is in a stable state, the difference between the tilt angle of the attitude at the current moment and the tilt angle of the attitude at the previous moment is calculated. Based on the comparison between the difference and the preset micro-motion threshold, it determines whether the image set correction matrix at the current moment needs to be updated.

[0183] The cache duration judgment unit is used to determine whether the cache duration of the image geometric correction matrix exceeds the maximum effective duration. If it exceeds the maximum effective duration, the image geometric correction matrix of the pose at the current time is reconstructed; if it does not exceed the maximum effective duration, the image geometric correction matrix cached at the previous time is used as the image geometric correction matrix at the current time.

[0184] The horizontal calibration unit is used to transform the currently acquired video image from the power monitoring and control sphere based on the image geometric correction matrix to obtain a virtual horizontal view image after horizontal calibration.

[0185] Example 4

[0186] In this embodiment, a virtual horizontal calibration device for power monitoring surveillance cameras is proposed. This device is used to implement a virtual horizontal calibration method for power monitoring surveillance cameras. A schematic diagram of its structure is shown below. Figure 4 As shown, it includes:

[0187] The data acquisition and preprocessing module is used to acquire real-time acceleration data in the three axes of the current power monitoring deployment ball, and to preprocess the acquired acceleration data.

[0188] The tilt angle calculation module is used to calculate the tilt angle of the current attitude of the power monitoring deployment ball based on the preprocessed acceleration data.

[0189] The image geometric correction matrix construction module is used to construct and cache the image geometric correction matrix of the power monitoring deployment ball at the current moment based on the tilt angle of the power monitoring deployment ball at the current moment.

[0190] The adaptive cache control module is used to calculate the variance of the tilt angle within the sliding window of the preset attitude frame at the current moment. Based on the comparison between the variance of the tilt angle and the preset stable state safety threshold, it determines whether the power monitoring deployment ball is in a stable state. If the power monitoring deployment ball is in an unstable state, the image geometric correction matrix cached at the previous moment is used as the image geometric correction matrix at the current moment. If the power monitoring deployment ball is in a stable state, the difference between the tilt angle of the attitude at the current moment and the tilt angle of the attitude at the previous moment is calculated. Based on the comparison between the difference and the preset micro-motion threshold, it determines whether the image set correction matrix at the current moment needs to be updated.

[0191] The cache duration judgment module is used to determine whether the cache duration of the image geometric correction matrix exceeds the maximum effective duration. If it exceeds the maximum effective duration, the image geometric correction matrix of the current pose is reconstructed; if it does not exceed the maximum effective duration, the image geometric correction matrix cached at the previous time step is used as the image geometric correction matrix of the current time step.

[0192] The horizontal calibration module is used to transform the currently acquired video image from the power monitoring sphere based on the image geometric correction matrix to obtain a virtual horizontal view image after horizontal calibration.

[0193] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for image virtual horizontal calibration of a surveillance ball for power monitoring, characterized in that, Includes the following steps: S1. Collect real-time acceleration data in the three axes of the current power monitoring deployment ball, and preprocess the collected acceleration data; S2. Based on the preprocessed acceleration data, calculate the tilt angle of the power monitoring and control ball at the current moment; S3. Based on the tilt angle of the power monitoring satellite at the current moment, construct and cache the image geometric correction matrix of the power monitoring satellite at the current moment, and execute S4; S4. Calculate the variance of the tilt angle within the preset frame sliding window at the current time, and determine whether the variance of the tilt angle is less than the preset stable state safety threshold. If yes, the power monitoring deployment ball is in a stable state, and proceed to S6; otherwise, the power monitoring deployment ball is in an unstable state, and proceed to step S5. S5. Use the image geometric correction matrix cached at the previous time step as the image geometric correction matrix at the current time step, and execute S7; S6. Calculate the difference between the tilt angle at the current moment and the tilt angle at the previous moment. If the difference is less than the preset micro-motion threshold, use the image geometric correction matrix cached at the previous moment as the image geometric correction matrix at the current moment and execute S7. Otherwise, return to S3. S7. Determine whether the cache duration of the image geometric correction matrix exceeds the maximum valid duration. If yes, reconstruct the image geometric correction matrix of the current pose and execute S8; otherwise, use the cached image geometric correction matrix of the previous time step as the image geometric correction matrix of the current time step and execute S8. S8. Based on the image geometric correction matrix, transform the currently acquired video image from the power monitoring satellite to obtain a virtual horizontal view image after horizontal calibration.

2. The image virtual horizontal calibration method for the power monitoring surveillance ball according to claim 1, wherein, The system collects real-time acceleration data along three axes from the current power monitoring satellite and preprocesses the collected acceleration data, including: Collecting raw acceleration data in three-axis directions of a current power monitoring surveillance ball ; Using a pre-calibrated alignment matrix The raw acceleration data is transformed to calculate the acceleration vector in the coordinate system of the power monitoring PTZ camera. The expression is: in, The components in the three-axis directions are represented as follows: , express Acceleration data in the axial direction, express Acceleration data in the axial direction, express Acceleration data in the axial direction.

3. The image virtual horizontal calibration method of the power monitoring surveillance ball according to claim 2, characterized in that, The tilt angle includes the pitch angle and the roll angle; The calculation of the tilt angle of the current attitude of the power monitoring and control ball includes: real-time calculation of the pitch angle and roll angle of the current attitude of the power monitoring and control ball, with the following expressions: in, This represents the pitch angle at a specific moment in the original video image from the power monitoring PTZ camera. This represents the roll angle at a specific moment in the original video image of the power monitoring and control sphere.

4. The image virtual horizontal calibration method for the power monitoring surveillance ball according to claim 3, characterized in that, The process of constructing and caching the image geometric correction matrix based on the current tilt angle of the power monitoring sphere includes: Constructing a matrix of tilt compensating rotors that are inversely tilted along the roll angle and inversely tilted along the pitch angle with the expression Matrix for tilt-compensated rotation Matrix for tilt-compensated rotation Matrix for tilt-compensated rotation is ； Let the horizontal resolution of the video images acquired by the power monitoring PTZ camera be . and image vertical resolution Based on horizontal resolution and image vertical resolution Obtain the camera parameter matrix of the power monitoring PTZ camera. The expression is: in, , , Construct the image geometric correction matrix The expression is: 。 5. The image virtual horizontal calibration method for the power monitoring surveillance ball according to claim 4, wherein, The calculation of the variance of the tilt angle within the preset frame sliding window at the current time includes: Using a sliding average filter, the average pitch and average roll angles are calculated for the frame interval of the sliding window, expressed as: θp = 1 / N ∑θp(i) for i = 1 to N wherein, represents the current frame index in the matrix smoothing transition process, represents a pre-set smoothing window; Based on the average pitch angle and the average roll angle, a variance of the pitch angle of the current attitude falling into the preset sliding window is calculated and the variance of the roll angle , respectively, and the expressions are as follows: in, The pitch angle represents the attitude at the current moment. This represents the deviation between the pitch angle and the average pitch angle at each moment under the current image geometric correction matrix. The roll angle represents the attitude at the current moment. This indicates the deviation between the roll angle and the average roll angle at each time step under the current image geometric correction matrix; the variance of the tilt angle is taken as the variance of the pitch angle the variance of the roll angle the maximum value in the variance of the roll angle 6. The image virtual horizontal calibration method for the power monitoring surveillance ball according to claim 5, wherein, calculating a difference between a tilt angle of the current time attitude and a tilt angle of the previous time attitude , the expression is: in, The pitch angle represents the attitude at the current moment. The pitch angle represents the attitude at the previous moment. The roll angle represents the attitude at the current moment. This indicates the roll angle of the attitude at the previous moment.

7. The image virtual horizontal calibration method for the power monitoring surveillance ball according to claim 6, wherein, When using the image geometric correction matrix cached at the previous time step as the image geometric correction matrix at the current time step, the method also includes using a matrix difference algorithm to smoothly transition the image geometric correction matrix from the previous time step to the image geometric correction matrix cached at the current time step, including: Pre-set the number of frames required for matrix transition , calculate the interpolation coefficients, expressed as: wherein represents the index of the current frame in the matrix smoothing transition process; Based on the interpolation coefficients, the current image geometric correction matrix, and the cached image geometric correction matrix from the previous time step, the image geometric correction matrix for the k-th frame during matrix transition is calculated. The expression is: wherein, denotes the image geometry correction matrix of the previous time instant, denotes the image geometry correction matrix of the current time instant.

8. The image virtual horizontal calibration method of the power monitoring surveillance ball according to claim 1, wherein, Also includes: Based on the horizontally calibrated virtual horizontal view image, the climbing behavior of the monitoring targets of the power monitoring sphere is analyzed, including: Perform target detection on the horizontally calibrated virtual horizontal viewpoint image to obtain the target bounding box; extracting a vertical coordinate of a bottom side of the target bounding box based on a preset pixel-length conversion factor and a horizontal calibrated virtual horizontal view image calculating a real height above sea level of the target mapped from the vertical coordinate ​ Determine whether the actual height above the ground exceeds a preset height threshold; if so, the monitored target is exhibiting climbing behavior. If not, judging the actual height from the ground If yes, the monitoring target has a climbing behavior. In other cases, the monitored target does not engage in any climbing behavior.

9. An image virtual level calibration system for a power monitoring surveillance ball, comprising: The system is used to implement the method according to any one of claims 1 to 8, comprising: The data acquisition and preprocessing unit is used to acquire real-time acceleration data in the three axes of the current power monitoring and control sphere, and to preprocess the acquired acceleration data. The tilt angle calculation unit is used to calculate the tilt angle of the current attitude of the power monitoring deployment ball based on the preprocessed acceleration data. The image geometric correction matrix construction unit is used to construct and cache the image geometric correction matrix of the power monitoring deployment ball at the current moment based on the tilt angle of the power monitoring deployment ball at the current moment. The adaptive cache control unit is used to calculate the variance of the tilt angle within the sliding window of the preset attitude frame at the current moment. Based on the comparison between the variance of the tilt angle and the preset stable state safety threshold, it determines whether the power monitoring deployment ball is in a stable state. If the power monitoring deployment ball is in an unstable state, the image geometric correction matrix cached at the previous moment is used as the image geometric correction matrix at the current moment. If the power monitoring deployment ball is in a stable state, the difference between the tilt angle of the attitude at the current moment and the tilt angle of the attitude at the previous moment is calculated. Based on the comparison between the difference and the preset micro-motion threshold, it determines whether the image set correction matrix at the current moment needs to be updated. The cache duration judgment unit is used to determine whether the cache duration of the image geometric correction matrix exceeds the maximum effective duration. If it exceeds the maximum effective duration, the image geometric correction matrix of the pose at the current time is reconstructed; if it does not exceed the maximum effective duration, the image geometric correction matrix cached at the previous time is used as the image geometric correction matrix at the current time. The horizontal calibration unit is used to transform the currently acquired video image from the power monitoring and control sphere based on the image geometric correction matrix to obtain a virtual horizontal view image after horizontal calibration.

10. A virtual horizontal calibration device for images of power monitoring surveillance cameras, characterized in that, The device is used to implement the method according to any one of claims 1 to 8, comprising: The data acquisition and preprocessing module is used to acquire real-time acceleration data in the three axes of the current power monitoring deployment ball, and to preprocess the acquired acceleration data. The tilt angle calculation module is used to calculate the tilt angle of the current attitude of the power monitoring deployment ball based on the preprocessed acceleration data. The image geometric correction matrix construction module is used to construct and cache the image geometric correction matrix of the power monitoring deployment ball at the current moment based on the tilt angle of the power monitoring deployment ball at the current moment. The adaptive cache control module is used to calculate the variance of the tilt angle within the sliding window of the preset attitude frame at the current moment. Based on the comparison between the variance of the tilt angle and the preset stable state safety threshold, it determines whether the power monitoring deployment ball is in a stable state. If the power monitoring deployment ball is in an unstable state, the image geometric correction matrix cached at the previous moment is used as the image geometric correction matrix at the current moment. If the power monitoring deployment ball is in a stable state, the difference between the tilt angle of the attitude at the current moment and the tilt angle of the attitude at the previous moment is calculated. Based on the comparison between the difference and the preset micro-motion threshold, it determines whether the image set correction matrix at the current moment needs to be updated. The cache duration judgment module is used to determine whether the cache duration of the image geometric correction matrix exceeds the maximum effective duration. If it exceeds the maximum effective duration, the image geometric correction matrix of the current pose is reconstructed; if it does not exceed the maximum effective duration, the image geometric correction matrix cached at the previous time step is used as the image geometric correction matrix of the current time step. A horizontal calibration module is configured to transform the video image currently collected by the power monitoring surveillance ball based on the image geometric correction matrix to obtain a virtual horizontal perspective image after horizontal calibration.

Images