Device and method for measuring, modeling and compensating imaging parameter variation of thermotropic camera

Abstract

The invention relates to the technical field of high-precision vision measurement and detection, and discloses a thermotropic camera imaging parameter change measurement, modeling and compensation device and method, and the method comprises the following steps: in a variable-temperature environment, synchronously collecting a series of image data and temperature data in one-to-one correspondence with the image data in collection time; processing the series of image data to solve a series of camera imaging parameters corresponding to the image data; the series of camera imaging parameters are associated with the series of temperature data by acquisition times. At least two calibration modules with different poses are fixedly deployed in a variable temperature environment, and a single-frame image global optimization algorithm is adopted to solve camera imaging parameters, so that the technical problems that a plurality of calibration plates with different positions need to be shot in a traditional calibration method, time is consumed, time is consumed, and transient changes of the parameters cannot be captured are solved; and the calibration efficiency and precision are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of high-precision visual measurement and detection technology, specifically to a device and method for measuring, modeling, and compensating for changes in thermal camera imaging parameters. Background Technology

[0002] In recent years, with the rapid development of many fields such as precision manufacturing, topographic mapping, and aerospace, photogrammetry has played an indispensable role. This technology, through processing and analyzing digital images acquired by cameras, can accurately obtain the geometric information of target objects. The measurement accuracy of a photogrammetry system directly determines its application effect and reliability. Practical research shows that the quality of digital images acquired by cameras is a key factor affecting measurement accuracy. High-quality images can provide clear and accurate raw data for subsequent measurement and analysis, helping to accurately reconstruct the characteristics and spatial relationships of target objects.

[0003] However, in practical applications, temperature variations in the imaging system significantly impact measurement accuracy, becoming a key issue restricting the application of photogrammetry in high-precision scenarios. With the application of modern high-precision photogrammetry systems in projects such as aerospace component dimensional inspection and high-precision terrain change monitoring, even millimeter-level or micrometer-level measurement errors can lead to drastically different results. Imaging systems often face complex temperature variations. In outdoor scenarios, seasonal changes and day-night cycles cause temperature fluctuations, and the heat generated during equipment operation alters the camera's structural temperature. These factors place the imaging system in a variable-temperature environment. Temperature changes alter the physical properties of internal optical and electronic components, leading to problems such as imaging temperature drift and inconsistent pixel response. Ultimately, this results in significant measurement errors, severely affecting the accuracy and reliability of measurement results and limiting the further application of this technology in high-precision fields.

[0004] In the field of photogrammetry, previous theoretical research has focused on camera geometric calibration and image processing algorithm optimization. Research on the imaging characteristics, internal mechanisms, and temperature-induced error compensation of imaging systems under different temperature environments is relatively insufficient. In terms of calibration methods, existing technologies generally lack high-precision and rapid calibration methods suitable for variable temperature environments. Traditional calibration methods (such as Zhang Zhengyou's calibration method) require taking multiple images of the calibration board at different locations at different times. This process is cumbersome and time-consuming, making it difficult to meet the real-time calibration requirements of cameras under variable temperature environments (such as spacecraft in-orbit fluctuations). It is also unable to capture the instantaneous changes in imaging parameters during dynamic temperature changes. Secondly, in terms of modeling accuracy, traditional mechanism modeling or structural thermodynamic analysis is difficult to accurately characterize the nonlinear changes in imaging parameters of cameras under thermal coupling deformation of multiple components (optical, electronic, mechanical). For example, the linear regression model mentioned in the study is not effective in compensating for complex nonlinear and thermal hysteresis effects, resulting in error compensation effects that cannot meet the accuracy requirements of high-precision photogrammetry systems (such as micron-level detection of aerospace components). Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a device and method for measuring, modeling, and compensating for changes in thermally induced camera imaging parameters. This solves the problem of the lack of a method in the prior art that can measure and compensate for camera imaging parameter drift caused by temperature changes (including nonlinearity, thermal inertia, and thermal hysteresis effects) in real time and with high precision.

[0006] To achieve the above objectives, the present invention provides a compensation method for measuring and modeling changes in thermally induced camera imaging parameters, comprising the following steps:

[0007] S1. Under varying temperature conditions, a series of image data and temperature data corresponding one-to-one with the image data in terms of acquisition time are simultaneously acquired.

[0008] S2. Process the series of image data to calculate a series of camera imaging parameters corresponding to the image data;

[0009] S3. Associate the series of camera imaging parameters with the series of temperature data according to the acquisition time to construct a time-series dataset, which is used to characterize the temporal relationship between the temperature data and the camera imaging parameters;

[0010] S4. A prediction model is trained based on the time-series dataset; real-time temperature data is acquired and input into the trained prediction model to output predicted camera imaging parameters; the predicted camera imaging parameters are used to compensate for thermal changes during the imaging process.

[0011] Preferably, the synchronous acquisition of a series of image data includes: fixing at least two calibration modules with different poses within the field of view of the imaging module, and acquiring a single frame image containing the at least two calibration modules as the image data.

[0012] Preferably, the acquired temperature data includes: synchronously acquired ambient temperature data, and at least one temperature data deployed at a key part of the imaging module.

[0013] Preferably, the process of processing a series of image data to calculate a series of camera imaging parameters corresponding to the image data includes:

[0014] The single-frame image is processed to extract the image plane coordinates of all target points in the at least two calibration modules;

[0015] Based on the image plane coordinates of all target points and their corresponding world coordinates, the camera imaging parameters are obtained iteratively through a global optimization algorithm.

[0016] Preferably, the camera imaging parameters include: principal distance, principal point coordinates, and at least one distortion coefficient.

[0017] Preferably, training a prediction model based on the time-series dataset includes: constructing a time-series prediction model, wherein the time-series prediction model is a recurrent neural network or a long short-term memory network (LSTM); and processing the time-series dataset into data within a predetermined time window. The temperature data sequence within the specified time window is used as input. The camera imaging parameters corresponding to the end time are used as the output training samples to train the time-series prediction model.

[0018] Preferably, acquiring real-time temperature data and inputting the real-time temperature data into the trained prediction model includes: acquiring real-time temperature data and constructing a time window of length [the specified value]. The temperature data sequence is then input into the time-series prediction model to output the predicted camera imaging parameters.

[0019] Preferably, the distortion coefficient is selected from at least one of radial distortion, tangential distortion, or affine distortion.

[0020] Preferably, the device for measuring, modeling, and compensating for changes in thermal camera imaging parameters includes:

[0021] Temperature control module, used to provide a variable temperature environment;

[0022] The imaging module is used to acquire image data;

[0023] At least one calibration module is positioned within the field of view of the imaging module;

[0024] At least one temperature sensor is used to collect temperature data;

[0025] A data processing module, connected to the imaging module and the at least one temperature sensor, is used for:

[0026] In the variable temperature environment, the imaging module and the at least one temperature sensor are controlled to synchronously acquire a series of image data and temperature data that corresponds one-to-one with the image data in terms of acquisition time.

[0027] The series of image data is processed to calculate a series of camera imaging parameters corresponding to the image data;

[0028] The series of camera imaging parameters and the series of temperature data are correlated according to the acquisition time to construct a time-series dataset;

[0029] A prediction model is trained based on the aforementioned time-series dataset;

[0030] Acquire real-time temperature data and input the real-time temperature data into the trained prediction model to output predicted camera imaging parameters;

[0031] The predicted camera imaging parameters are used to compensate for thermal changes during the imaging process.

[0032] Preferably, a data processing apparatus includes a processor and a memory, the memory storing a computer program, the processor being configured to, when executing the computer program, implement the method as described in any one of claims 1 to 8.

[0033] This invention provides a device and method for measuring, modeling, and compensating for changes in imaging parameters of a thermally induced camera. It offers the following advantages:

[0034] 1. This invention solves the technical problems of traditional calibration methods, which require taking multiple images of calibration boards at different positions, are time-consuming and cumbersome, and cannot capture instantaneous parameter changes, by using at least two fixedly deployed calibration modules in a variable temperature environment and employing a single-frame image global optimization algorithm to solve camera imaging parameters. This method can reduce calibration time from several hours to minutes, and the root mean square error (RMSE) of reprojection is as low as 1 / 25.461 pixels, significantly improving calibration efficiency and accuracy and meeting the real-time calibration requirements in a variable temperature environment.

[0035] 2. This invention constructs a time-series dataset between temperature data and camera imaging parameters, and trains it using a Long Short-Term Memory (LSTM) network, enabling it to learn and capture the time-series dependency and thermal hysteresis effect between temperature changes and imaging parameters. This method solves the technical problem that traditional mechanism modeling or thermodynamic analysis is difficult to accurately characterize the nonlinear change characteristics under thermal coupling deformation of multiple components, and achieves high-precision model prediction.

[0036] 3. In the process of constructing the time series dataset, the present invention preferably uses the wavelet threshold denoising method to process the collected raw data. Compared with the traditional mean filtering, this method can effectively filter out noise while retaining more than 90% of the temperature-related data details. This provides a high-quality data foundation for the subsequent training of the LSTM model, ensuring that the model can learn the subtle parameter change rules, thereby improving the robustness of the prediction model and the final compensation accuracy.

[0037] 4. This invention acquires real-time temperature data and inputs it into a trained prediction model to output predicted camera imaging parameters in real time and perform compensation. This achieves dynamic suppression of thermally induced errors in practical applications. This method does not require modification of existing camera hardware. Temperature error compensation can be achieved simply by setting up an experimental environment and optimizing the algorithm. It has significant engineering application value and wide applicability.

[0038] 5. This invention constructs a multi-dimensional temperature input vector by simultaneously collecting ambient temperature and at least one temperature data deployed in key parts of the imaging module under dynamic temperature variations. This method enables the prediction model to not only learn the influence of ambient temperature, but also to learn the complex nonlinear response of internal heat sources (such as heat generation of electronic components) and thermal coupling deformation of multiple components, thereby more comprehensively characterizing the true thermal state of the camera. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating the system implementation of the present invention;

[0040] Figure 2 This is a schematic diagram of the original experimental image and target point division for this invention;

[0041] Figure 3 This is a schematic diagram of the reprojection error field results of the present invention;

[0042] Figure 4 This is a schematic diagram of the experimental temperature data and imaging parameters of the present invention.

[0043] Figure 5 This is a schematic diagram illustrating the relationship between imaging parameters and temperature changes in this invention;

[0044] Figure 6 This is a diagram showing the predicted imaging parameters of the present invention;

[0045] Figure 7 This is a diagram showing the error distribution of the imaging parameters of the present invention;

[0046] Figure 8 This is one of the method flowcharts of the present invention;

[0047] Figure 9 This is the second flowchart of the method of the present invention. Detailed Implementation

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Please see the appendix Figure 8 -Appendix Figure 9 This invention provides a compensation method for measuring and modeling changes in imaging parameters of a thermally induced camera, comprising the following steps:

[0050] S1. In a variable temperature environment, a series of image data and temperature data corresponding one-to-one with the image data are simultaneously acquired. The simultaneous acquisition of the series of image data includes: at least two calibration modules with different poses are fixedly deployed within the field of view of the imaging module, and a single frame image containing at least two calibration modules is acquired as image data. The series of image data is processed to calculate a series of camera imaging parameters corresponding to the image data, including: processing the single frame image to extract the image plane coordinates of all target points in at least two calibration modules; and based on the image plane coordinates of all target points and their corresponding world coordinates, the camera imaging parameters are iteratively solved through a global optimization algorithm. The camera imaging parameters include: principal distance, principal point coordinates, and at least one distortion coefficient, which is selected from at least one of radial distortion, tangential distortion, or affine distortion. The acquisition of temperature data includes: simultaneously acquiring ambient temperature data and at least one temperature data deployed at key parts of the imaging module.

[0051] Specifically, in the photogrammetric camera imaging model, the first... The first calibration board Spatial coordinates of the target point To the coordinates of the target point in the image The projection process follows the camera imaging mathematical model constructed based on the collinearity equation, and the specific relationship is shown in equation (1).

[0052]

[0053] In the formula, The camera's principal distance; For the camera in the Attitude matrix in the calibration plate coordinate system For the camera in the Relative spatial position in the coordinate system of the calibration plate; For the first The first calibration board Image plane coordinate imaging distortion of a spatial target point; The coordinates of the camera's principal point; Radial distortion coefficient; The tangential distortion coefficient; The affine distortion coefficient; For the first The first calibration board The distance between the image plane coordinates of a spatial target point and the principal point coordinates.

[0054] For any spatial point coordinates The projection equation (1) is expanded using Taylor's formula to obtain a linear approximation equation, as shown in equation (2).

[0055]

[0056] In the formula This indicates that the image plane coordinates are calculated using the back intersection algorithm. These are the image plane coordinates of the actual target points. Construct all... In the calibration board The large extended error equation for the coordinates of a spatial target point is shown in equation (3).

[0057]

[0058] in, The Jacobian matrix represents the camera's exterior orientation parameters. The Jacobian matrix represents the camera's interior orientation parameters. To solve the first Camera exterior orientation parameters in a calibration plate coordinate system; For the camera's internal orientation parameters (principal point, principal distance, distortion coefficient), equation (3) can be implicitly expressed as equation (4).

[0059]

[0060] Then the Jacobian matrix Normalization is performed to eliminate scale differences between different parameter dimensions, improving the stability and convergence efficiency of subsequent optimization algorithms. The Jacobian matrix is ​​then completed. After normalization, LU decomposition is used, and then the LM iterative algorithm is used to iteratively correct the initial estimates of all parameters until the camera interior orientation parameters, distortion coefficients and camera exterior orientation parameters solved in all calibration plate coordinate systems reach the global optimum.

[0061] S2. Process a series of image data to calculate a series of camera imaging parameters corresponding to the image data;

[0062] S3. Associate a series of camera imaging parameters with a series of temperature data according to the acquisition time to construct a time-series dataset. The time-series dataset is used to characterize the temporal relationship between temperature data and camera imaging parameters.

[0063] S4. A prediction model is trained based on a time-series dataset; real-time temperature data is acquired and input into the trained prediction model to output predicted camera imaging parameters; the predicted camera imaging parameters are used to compensate for thermal changes during the imaging process. Acquiring real-time temperature data and inputting it into the trained prediction model includes: acquiring real-time temperature data and constructing a time window of length [missing information]. The temperature data sequence is used as the basis for training a prediction model. This model is then fed into a time-series prediction model to output predicted camera imaging parameters. The process involves: constructing a time-series prediction model, which can be a recurrent neural network or a long short-term memory network (LSTM); and processing the time-series dataset into parameters within a predetermined time window. The temperature data sequence within the time window is used as input. The camera imaging parameters corresponding to the end time are used as the output training samples to train the time series prediction model.

[0064] Please see the appendix Figure 1 A device for measuring, modeling, and compensating for changes in imaging parameters of a thermally induced camera, including:

[0065] Temperature control module, used to provide a variable temperature environment;

[0066] The imaging module is used to acquire image data;

[0067] At least one calibration module is positioned within the field of view of the imaging module;

[0068] At least one temperature sensor is used to collect temperature data;

[0069] The data processing module, connected to the imaging module and at least one temperature sensor, is used for:

[0070] In a variable temperature environment, the control imaging module and at least one temperature sensor synchronously acquire a series of image data and temperature data that corresponds one-to-one with the image data in terms of acquisition time.

[0071] A series of image data are processed to calculate a series of camera imaging parameters corresponding to the image data;

[0072] A series of camera imaging parameters are correlated with a series of temperature data according to the acquisition time to construct a time-series dataset;

[0073] A prediction model is trained based on a time-series dataset;

[0074] Acquire real-time temperature data and input the real-time temperature data into the trained prediction model to output the predicted camera imaging parameters;

[0075] Predicted camera imaging parameters are used to compensate for thermal changes during the imaging process;

[0076] Specifically, firstly, an image for imaging parameter calibration is acquired. After image processing, the center phase plane coordinates of the target point are obtained. Then, the target points are divided and matched. Combined with the world coordinates of the target points, the back intersection algorithm is used to obtain the initial values ​​of the camera imaging parameters. Then, the high-precision camera intrinsic and extrinsic parameters are solved iteratively using the LM algorithm. Next, the acquired temperature data is aligned with the imaging parameters to obtain the temperature-imaging parameter dataset. Subsequently, the data is preprocessed, and the input model is trained and optimized to finally obtain the predicted camera imaging parameters.

[0077] (1) System assembly

[0078] Construction of fixed support structure: First, fix the carbon fiber bracket to the bottom of the variable temperature test chamber, and then install the camera bracket and calibration plate bracket through Invar alloy connectors to ensure that the bracket is installed firmly and without shaking, so as to avoid structural displacement affecting the data accuracy in subsequent experiments.

[0079] Imaging module installation and debugging: Assemble the industrial camera and optical lens, fix them on the camera bracket, adjust the camera position and lens focal length, and confirm through the camera preview function that all 5 calibration boards are within the field of view and the image is clear. Record the initial installation parameters of the camera and lens at this time.

[0080] Calibration module deployment: Fix the 5 calibration boards one by one to the calibration board bracket, adjust the position and pose of each calibration board to ensure that there is no obstruction between adjacent calibration boards, turn on the light equalization plate, and collect a single frame test image through the computer to check whether the brightness of the target point is uniform (grayscale value fluctuation ≤10). If it is not uniform, adjust the position or brightness of the light equalization plate.

[0081] Temperature acquisition module installation test: Install the 6-channel temperature sensor probes to the designated monitoring point, connect the data acquisition instrument to the industrial computer, and observe whether the sensor values ​​are stable after powering on (fluctuation ≤0.1℃ within 10 minutes). If the values ​​jump, check whether the sensor contact is good or replace the probes.

[0082] System integration and testing: Start the camera and temperature acquisition device, set the acquisition frequency to 1Hz, and simultaneously acquire 100 frames of images and corresponding temperature data to verify whether the data transmission is smooth and whether the time and space are aligned, and ensure that the overall system functions normally.

[0083] (2) Data collection for calibration experiments

[0084] Temperature program setting: According to the experimental requirements, the temperature cycle program of "40℃-0℃-40℃" is set through the variable temperature test chamber control system. The temperature drops from 40℃ to 0℃ and then rises back to 40℃ to complete one temperature cycle. A total of 3 cycles are run to ensure that sufficient multi-temperature data are obtained.

[0085] (3) Processing of raw image and temperature data

[0086] Target point coordinate extraction and segmentation matching: For the original image, such as Figure 2 As shown in (a), Gaussian filtering and binarization are performed to obtain the preprocessed image. Then, an ellipse fitting algorithm is used to extract the image plane center coordinates of the circular target points on the calibration board. Target points on different calibration boards are divided according to their spatial positions and matched with known world coordinates of the target points to ensure that each image plane coordinate corresponds to a unique world coordinate. The results are as follows: Figure 2 As shown in (b).

[0087] Imaging parameter calibration: Combining the world coordinates and image plane coordinates of the matched target points, the back intersection algorithm is used to calculate the initial values ​​of the camera's intrinsic and extrinsic parameters without distortion coefficients. The initial values ​​are substituted into the aforementioned single-frame camera imaging parameter calibration method under varying temperature conditions to construct a large-scale extended error equation for all target points on the calibration board. The Jacobian matrix is ​​normalized to eliminate parameter dimensional differences. Then, the parameters are iteratively corrected using LU decomposition and LM iteration algorithms until the globally optimal camera intrinsic parameters (principal distance, principal point coordinates, distortion coefficients) and extrinsic parameters are solved, thus completing the imaging parameter calibration.

[0088] Temperature and imaging parameter alignment: The real-time temperature data collected by the 6 temperature sensors is associated with the imaging parameters of the corresponding timestamps. Abnormal data (such as imaging parameter calibration failures and data with temperature jumps) are removed to form a time-series temperature-imaging parameter dataset, which is divided into training set, validation set and test set in a 7:2:1 ratio to prepare for subsequent LSTM model training.

[0089] (4) Training of LSTM network model

[0090] LSTM network input and output: The input is temperature data for 20 consecutive time steps (ambient temperature + temperatures of 5 key camera components), with a dimension of (20,6), covering the cumulative temperature effect and thermal response delay. The output is the predicted values ​​of 9 camera imaging parameters (principal distance f; principal points x0, y0; radial distortion k1, k2, k3; tangential distortion p1, p2; affine distortion b1, b2), used to compensate for the original calibration results.

[0091] Model Training and Optimization: Based on the aforementioned LSTM-based camera imaging parameter prediction method, using the root mean square error (RMSE) as the loss function, the Adam optimizer (learning rate 0.001, batch size 32, iterations 100) is employed. The model is trained using the training set, and its performance is evaluated using the validation set every 10 iterations (by calculating the validation set imaging parameter prediction RMSE). If the validation set RMSE does not decrease for 5 consecutive iterations, training is stopped to avoid overfitting. During training, the model's prediction performance for parameters in different temperature ranges is monitored in real time to ensure its adaptability across the entire temperature range.

[0092] (5) Parameter compensation

[0093] LSTM network input and output: The input is temperature data for 10 consecutive time steps (6 temperature features in total), with a dimension of (10,6). By capturing the temporal features and cumulative effects of temperature changes, it reflects the dynamic influence of temperature on camera parameters. The output is the compensated predicted values ​​of 10 camera imaging parameters, including principal distance f, principal point x0, principal point y0, radial distortion k1, k2, k3, tangential distortion p1, p2, and affine distortion b1, b2, with a dimension of (10), which are used to correct the original calibration results to offset temperature interference.

[0094] Data Input and Range Selection: The preprocessed temperature sequence data and corresponding camera parameter labels are loaded from the file. The range of parameters to be compensated is determined by obtaining the start and end sequence indices input by the user. The input index must meet validity checks (non-negative start index, end index less than the total data length, and start index not greater than the end index). If invalid, the user is prompted to re-enter the data.

[0095] Model application process: Load the pre-trained LSTM model (the model contains a 2-layer LSTM network, a hidden layer size of 64, and a Dropout rate of 0.2) and set it to evaluation mode; perform denormalization on the temperature sequence within the user-specified index range, input it into the model inference to obtain the parameter prediction values, and then convert them into actual physical parameter values ​​through denormalization.

[0096] Results verification: By visually comparing the predicted and actual values ​​of 10 parameters, a 5×2 subplot layout is used to display the changing trends of each parameter, intuitively assessing the accuracy of the compensation effect; at the same time, the compensation accuracy can be quantified by calculating the absolute error between the predicted and actual values.

[0097] The experimental results are as follows:

[0098] The target point is reprojected and compared with the actual point. The reprojection error field is as follows: Figure 3 As shown in (a), the error statistics for the reprojection x-axis and y-axis are as follows: Figure 3 As shown in (b), the average error along the x-axis is -7.00481 × 10⁻⁶. -8 Average error along the y-axis: 2.9482 × 10 -8 x-axis standard deviation: 1.6366 × 10 -4 y-axis standard deviation: 1.5685 × 10 -4 The reprojection RMSE was 1 / 25.461 pixels. During the experiment, the temperature was set to decrease from 40℃ to 0℃ and then increase back to 40℃ as a temperature cycle. Temperature data for different parts of the camera were recorded (1. Ambient; 2. Lens front; 3. Lens rear; 4. Camera front; 5. Camera side; 6. Camera rear). Figure 4 As shown in (a), the temperature of various parts of the camera changes along with the ambient temperature, and the correlation between the camera imaging parameters and each temperature is as follows: Figure 4 As shown in (b), the imaging parameters The absolute value of the correlation coefficient with temperature is greater than 0.9, indicating that there is a strong correlation between each imaging parameter and temperature, and that temperature changes have a significant impact on the imaging parameters.

[0099] Visualize the various imaging parameters and temperature data in a time series, such as Figure 5 As shown, principal distance It exhibits a significant lag relative to temperature changes, while the main point During the cooling process at 15℃, the offset increased as the temperature decreased. Preliminary analysis suggests that the temperature change range exceeded the linear variation limit of the mechanical material in the x-axis direction. Therefore, the deviation of the principal point in the x-axis direction increased at both excessively high and low temperatures. The predicted and true values ​​of the imaging parameters were compared, and the visualization results are as follows: Figure 6 As shown, the predicted curves for most parameters closely match the actual curves, indicating that the model can capture the overall pattern of parameter changes with the sample sequence and performs well in trend prediction. The fact that the predicted curves almost overlap with the actual curves demonstrates the model's strong adaptability to slowly changing, periodically varying parameters. These parameters exhibit regular fluctuations due to environmental factors such as temperature, and the model effectively reproduces the change pattern by learning from historical sequences. The error distribution of the imaging parameters is shown in the figure. Figure 7 As shown.

[0100] The final RMSE results for each imaging parameter prediction are shown in the table below:

[0101]

[0102] The experimental results and accuracy analysis show that after image preprocessing, the target point center coordinates are extracted via ellipse fitting, and the camera intrinsic and extrinsic parameters and distortion coefficients are solved by the single-frame imaging parameter calibration algorithm. The average error of the reprojection x-axis is -7.00481 × 10⁻⁶. -8 The average error along the y-axis is 2.9482 × 10⁻⁶. -8 The RMSE reached 1 / 25.461 pixel, demonstrating high calibration accuracy. Within the experimental temperature range, the temperature of various parts of the camera changed synchronously with the environment, and the absolute value of the correlation coefficient between imaging parameters and temperature exceeded 0.9, confirming the significant impact of temperature on imaging parameters. Furthermore, the principal distance exhibited a lag with temperature changes, and the principal point offset was abnormal in specific temperature regions. In imaging parameter prediction, most predicted curves closely matched the actual curves, with the lowest predicted RMSE reaching 1.3762 × 10⁻⁶ pixels. -9 The model can accurately capture the pattern of parameter changes, fully verifying the effectiveness of the invention in calibration accuracy, temperature correlation analysis and parameter prediction.

[0103] Please see the appendix Figure 1 -Appendix Figure 9 A data processing apparatus includes a processor and a memory, wherein a computer program is stored in the memory, and the processor is configured to execute the computer program to implement the method as claimed in any one of claims 1 to 8;

[0104] Specifically, a nonlinear and dynamic mapping relationship between historical temperature data and camera parameter drift was established by using a time-series prediction model (such as LSTM). This allows the system to predict and compensate for parameter changes caused by thermal inertia and thermal hysteresis effects simply by monitoring the real-time temperature, thus solving the accuracy problem caused by thermal changes in parameters in variable temperature environments.

[0105] In summary, this invention provides a device and method for measuring, modeling, and compensating for changes in thermally induced camera imaging parameters. By utilizing at least two calibration modules with fixed deployments in different poses under varying temperature conditions and employing a single-frame image global optimization algorithm to solve the camera imaging parameters, it solves the technical problems of traditional calibration methods, which require taking multiple images of calibration boards at different positions, are time-consuming and cumbersome, and cannot capture instantaneous parameter changes. This significantly improves calibration efficiency and accuracy, meeting the real-time calibration requirements under varying temperature conditions. Furthermore, by constructing a time-series dataset between temperature data and camera imaging parameters and training it using a Long Short-Term Memory (LSTM) network, the system can learn and capture the time-series dependency and thermal hysteresis effect between temperature changes and imaging parameters, achieving high-precision model prediction.

[0106] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A compensation method for measuring and modeling changes in imaging parameters of a thermally induced camera, characterized in that, Includes the following steps: S1. Under varying temperature conditions, a series of image data and temperature data corresponding one-to-one with the image data in terms of acquisition time are simultaneously acquired. S2. Process the series of image data to calculate a series of camera imaging parameters corresponding to the image data; S3. Associate the series of camera imaging parameters with the series of temperature data according to the acquisition time to construct a time-series dataset, which is used to characterize the temporal relationship between the temperature data and the camera imaging parameters; S4. A prediction model is trained based on the time-series dataset; Real-time temperature data is acquired and input into the trained prediction model to output predicted camera imaging parameters; the predicted camera imaging parameters are then used to compensate for thermal changes during the imaging process.

2. The compensation method for measuring and modeling changes in thermally induced camera imaging parameters according to claim 1, characterized in that, The synchronous acquisition of a series of image data includes: fixing at least two calibration modules with different poses within the field of view of the imaging module, and acquiring a single frame image containing the at least two calibration modules as the image data.

3. The compensation method for measuring and modeling changes in thermally induced camera imaging parameters according to claim 1, characterized in that, The acquired temperature data includes: synchronously acquired ambient temperature data, and at least one temperature data point deployed at a key location of the imaging module.

4. The compensation method for measuring and modeling changes in thermally induced camera imaging parameters according to claim 2, characterized in that, The process of processing a series of image data to calculate a series of camera imaging parameters corresponding to the image data includes: The single-frame image is processed to extract the image plane coordinates of all target points in the at least two calibration modules; Based on the image plane coordinates of all target points and their corresponding world coordinates, the camera imaging parameters are obtained iteratively through a global optimization algorithm.

5. The compensation method for measuring and modeling changes in thermally induced camera imaging parameters according to claim 4, characterized in that, The camera imaging parameters include: principal distance, principal point coordinates, and at least one distortion coefficient.

6. The compensation method for measuring and modeling changes in thermally induced camera imaging parameters according to claim 1, characterized in that, The step of training a prediction model based on a time-series dataset includes: constructing a time-series prediction model, wherein the time-series prediction model is a recurrent neural network or a long short-term memory network (LSTM); and processing the time-series dataset into data with predetermined time windows. The temperature data sequence within the specified time window is used as input. The camera imaging parameters corresponding to the end time are used as the output training samples to train the time-series prediction model.

7. The compensation method for measuring and modeling changes in thermally induced camera imaging parameters according to claim 1, characterized in that, The step of acquiring real-time temperature data and inputting the real-time temperature data into the trained prediction model includes: acquiring real-time temperature data and constructing a time window of length [the specified time window]. The temperature data sequence is then input into the time-series prediction model to output the predicted camera imaging parameters.

8. The compensation method for measuring and modeling changes in thermally induced camera imaging parameters according to claim 5, characterized in that, The distortion coefficient is selected from at least one of radial distortion, tangential distortion, or affine distortion.

9. A device for measuring, modeling, and compensating for changes in imaging parameters of a thermally induced camera, characterized in that, The compensation method for measuring and modeling changes in thermally induced camera imaging parameters according to any one of claims 1-8 includes: Temperature control module, used to provide a variable temperature environment; The imaging module is used to acquire image data; At least one calibration module is positioned within the field of view of the imaging module; At least one temperature sensor is used to collect temperature data; A data processing module, connected to the imaging module and the at least one temperature sensor, is used for: In the variable temperature environment, the imaging module and the at least one temperature sensor are controlled to synchronously acquire a series of image data and temperature data that corresponds one-to-one with the image data in terms of acquisition time. The series of image data is processed to calculate a series of camera imaging parameters corresponding to the image data; The series of camera imaging parameters and the series of temperature data are correlated according to the acquisition time to construct a time-series dataset; A prediction model is trained based on the aforementioned time-series dataset; Acquire real-time temperature data and input the real-time temperature data into the trained prediction model to output predicted camera imaging parameters; The predicted camera imaging parameters are used to compensate for thermal changes during the imaging process.

10. A data processing apparatus, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor is configured to, when executing the computer program, implement the method as described in any one of claims 1 to 8.

Images