A method and system for ultraviolet-visible light image registration

Abstract

The application discloses a kind of ultraviolet-visible light image registration method and system, belong to image processing technical field.Method includes: establishing double state imaging system, the relative installation position of fixed ultraviolet detector component and visible light detector component;In first imaging state, establish basic registration relationship based on high signal-to-noise ratio ultraviolet image and visible light image;In second imaging state, establish filter state correction relationship by a small amount of ultraviolet LED control point;The basic registration relationship and state correction relationship are compounded, and the final registration relationship suitable for actual solar-blind ultraviolet detection working condition is obtained.The application stores basic registration and state correction decoupling, avoids direct complete calibration under low signal-to-noise ratio condition, and only needs a small amount of control point to complete state correction, significantly reduces engineering deployment cost, improves the stability and automation level of ultraviolet-visible light registration in multi-distance scene.

Description

Technical Field

[0001] This invention relates to the field of image processing technology; and more particularly to an ultraviolet-visible image registration method and system. Background Technology

[0002] Multispectral imaging technology, with its complementary fusion advantages of multi-channel information, is widely used in industrial inspection, medical diagnosis, and other fields, effectively compensating for the sensing limitations of single sensors in complex environments. Its implementation methods include single-detector multi-channel schemes such as single-detector spectral splitting, filter switching, and focal plane filter integration, as well as multi-view imaging schemes where multiple heterogeneous detectors acquire images of different spectral bands. When using multi-view heterogeneous detectors for cross-modal joint imaging (ultraviolet, visible, infrared, etc.), since each imaging channel typically has an independent optical path and detector plane, spatial coordinate inconsistencies often exist between the output images of different channels due to factors such as installation pose, optical field of view, imaging resolution, distortion characteristics, and differences in spectral response. Therefore, image registration is fundamental to achieving multi-view heterogeneous image fusion and target localization. Its core lies in establishing the spatial correspondence between the output images of different imaging channels to ensure the spatial consistency of the subsequent fusion results.

[0003] Existing image registration methods are mostly based on feature extraction and matching, establishing correspondences by detecting key points, edges, or region descriptors in the image. However, these methods are highly dependent on the richness of image features, requiring feature traversal of all image frames in dynamically changing scenes, resulting in high computational cost and poor real-time performance. Especially in cross-modal registration scenarios, such as ultraviolet and visible light images, or infrared and visible light images, due to significant differences in imaging mechanisms, the feature correlation between different modal images is weak, making feature mismatch or matching failures highly likely.

[0004] In the field of optoelectronic detection, ultraviolet imaging typically focuses on the 200–400 nm wavelength band. When ultraviolet light passes through the Earth's atmosphere, it is strongly absorbed and scattered by components such as oxygen, ozone, and water vapor. Ultraviolet radiation with wavelengths below 280 nm can hardly reach the Earth's surface, forming a natural "solar-blind" region. Solar-blind ultraviolet detectors operate in the 200–280 nm band, where solar background radiation is extremely weak, resulting in a very low false alarm rate and strong anti-interference capabilities, making them valuable for applications such as electrical discharge detection. However, due to the limited information and low image contrast in the solar-blind band, practical applications usually require collaboration with visible light detectors to achieve accurate target identification and localization through image fusion.

[0005] Existing methods for registering solar-blind ultraviolet (UV) and visible light images mainly rely on two technical approaches: one is to extract the centroid of bright spots as calibration points by superimposing a large number of solar-blind UV images and performing noise reduction filtering; the other is to use a UV light source to illuminate an open-aperture checkerboard target and utilize the corner features generated by the transmission aperture to complete calibration and registration. Both methods have significant limitations: UV detectors require a narrowband filter to be loaded at the front of the lens to limit the response band, resulting in a significant reduction in light transmittance, poor image signal-to-noise ratio, low brightness, weak contrast, and unstable corner feature extraction, leading to low calibration accuracy and insufficient registration reliability. Furthermore, the loading of the filter may alter the imaging optical path, introducing additional optical path difference and refraction deviation. Existing methods do not adequately consider independent modeling and compensation for these imaging state differences. In long-distance or multi-distance detection scenarios, the adaptability of registration accuracy to distance variations is particularly insufficient, failing to support the automated target detection and localization requirements under multi-distance conditions. Summary of the Invention

[0006] Based on the above analysis, the present invention provides an ultraviolet-visible light image registration method and system; solving the problems of low image brightness, poor signal-to-noise ratio, difficulty in stable extraction of calibration features, and difficulty in directly applying the calibration results without the filter to the actual working optical path after the filter is loaded, caused by loading a narrow-band solar-blind ultraviolet filter in the actual detection process.

[0007] This invention discloses a method for ultraviolet-visible image registration, comprising: A dual-state imaging system is established, with the relative installation positions of the ultraviolet detector assembly and the visible light detector assembly fixed, and the first imaging state and the second imaging state of the ultraviolet detector are set; the first imaging state is the state without a solar-blind ultraviolet filter, and the second imaging state is the state with a solar-blind ultraviolet filter. In the first imaging state, a basic registration relationship from ultraviolet image to visible light image is established based on the acquired ultraviolet image and visible light image; While keeping the relative positions of the two detectors unchanged, switch to the second imaging state. Based on the corresponding control points between the solar-blind ultraviolet image in the second imaging state and the ultraviolet image in the first imaging state, establish a state correction relationship from the second imaging state to the first imaging state. The basic registration relationship is combined with the state correction relationship to obtain the final registration relationship from the solar-blind ultraviolet image to the visible light image in the second imaging state.

[0008] Furthermore, the state correction relationship is a coordinate mapping relationship, which uniformly represents the equivalent focal length change, principal point offset, residual distortion, image rotation, scale change and local coordinate deviation caused by the loading of the filter as a two-dimensional coordinate transformation, and directly establishes the mapping between the ultraviolet image coordinate system of the second imaging state and the first imaging state.

[0009] Furthermore, the state correction relationship is represented by an affine transformation model, and the affine transformation matrix is ​​a 2×3 matrix. Its linear transformation submatrix Used to comprehensively describe rotation, scaling, shearing components, and translation components. Used to describe the overall image offset.

[0010] Furthermore, the number of corresponding control points is no less than three groups, and each control point is not collinear with the others; the state correction relationship is solved directly by the least squares method to obtain the affine transformation matrix parameters, or when there is outlier noise at the control points, a robust estimation solution is obtained by using a random sampling consensus algorithm.

[0011] Furthermore, in the acquisition of the state correction relationship, an LED dot matrix target composed of several ultraviolet LED point light sources arranged in a known geometric pattern is used as the control point carrier; the central radiation wavelength of the ultraviolet LED point light source is located within the passband of the solar-blind ultraviolet filter or the effective response band of the solar-blind ultraviolet detector assembly; in the first imaging state and the second imaging state, the centroid of the same group of LED light-emitting points is extracted to obtain the coordinates of the corresponding control point, and the target position and the relative pose of the two detectors remain unchanged during the acquisition process.

[0012] Furthermore, it also includes: establishing a multi-distance registration parameter library, and performing the steps of obtaining the basic registration relationship, obtaining the state correction relationship, and combining the registration relationship at multiple representative detection distances to obtain the basic registration relationship, state correction relationship, and final registration relationship corresponding to each distance; during actual detection, searching or interpolating in the multi-distance registration parameter library according to the current detection distance.

[0013] Furthermore, the interpolation call includes: if the current detection distance matches a certain calibrated distance, the corresponding final registration relationship is directly called; if the current detection distance is between two calibrated distances, the parameters of the basic registration relationship and the state correction relationship are interpolated separately and then combined to obtain the final registration relationship under the current detection distance.

[0014] The present invention also discloses an ultraviolet-visible image registration device, comprising: Rigid installation platform; An ultraviolet detector assembly, fixed on the rigid mounting platform, includes a solar-blind ultraviolet filter and a filter switching mechanism, wherein the filter switching mechanism is configured to switch between a first imaging state without the solar-blind ultraviolet filter and a second imaging state with the solar-blind ultraviolet filter loaded. The visible light detector assembly is fixed on the rigid mounting platform, maintaining a constant relative orientation with the ultraviolet detector assembly; The processor, which is communicatively connected to the ultraviolet detector assembly and the visible light detector assembly, is configured to: In the first imaging state, a basic registration relationship from ultraviolet image to visible light image is established based on the acquired ultraviolet image and visible light image; While keeping the relative positions of the two detectors unchanged, switch to the second imaging state. Based on the corresponding control points between the solar-blind ultraviolet image in the second imaging state and the ultraviolet image in the first imaging state, establish a state correction relationship from the second imaging state to the first imaging state. The state correction relationship uniformly represents the equivalent focal length change, principal point offset, residual distortion, image rotation, scale change and local coordinate deviation caused by filter loading as a two-dimensional coordinate transformation. The basic registration relationship is combined with the state correction relationship to obtain the final registration relationship from the solar-blind ultraviolet image to the visible light image in the second imaging state.

[0015] In the first imaging state, a basic registration relationship from ultraviolet image to visible light image is established based on the acquired ultraviolet image and visible light image; The basic registration relationship is combined with the state correction relationship to obtain the final registration relationship from the solar-blind ultraviolet image to the visible light image in the second imaging state.

[0016] Furthermore, it also includes target components. The target assembly includes a checkerboard target and an LED dot matrix target. The checkerboard calibration target is used for calibrating imaging parameters and establishing basic registration relationships in the first imaging state. The LED dot matrix target is composed of several ultraviolet LED point light sources arranged in a known geometric pattern, and is used to establish the state correction relationship in the second imaging state.

[0017] Furthermore, the processor is also configured to: A multi-distance registration parameter library is established. The establishment of the basic registration relationship, the establishment of the state correction relationship, and the combination of the final registration relationship are performed at multiple representative detection distances to obtain the basic registration relationship, state correction relationship, and final registration relationship corresponding to each distance. The current detection distance is used to search or interpolate the multi-distance registration parameter library.

[0018] Compared with traditional methods, the present invention has the following technical advantages: This invention does not perform complete UV-Vis calibration directly under the condition of loading a solar-blind UV filter. Instead, it first obtains a stable basic registration relationship using the unfiltered state, and then obtains a state correction relationship using the filtered state. This avoids the instability problem caused by direct complete calibration in low signal-to-noise ratio UV images after loading a solar-blind UV filter.

[0019] This invention compensates for possible changes in optical path, image shift, scale change, or local residuals caused by loading a solar-blind ultraviolet filter through a state correction relationship, so that the basic registration relationship obtained in the filterless state can be applied to the actual detection conditions after loading the filter.

[0020] This invention decouples and stores the "basic registration relationship" and the "state correction relationship", which can establish a registration parameter library at different detection distances or different focusing states, thereby improving the automation and engineering adaptability of ultraviolet-visible light registration in multi-distance scenarios.

[0021] The state correction process of this invention only requires a small number of corresponding control points. Compared with recalibrating under the condition of loading a solar-blind ultraviolet filter, it has the advantages of simplified operation, convenient parameter updates and low engineering deployment costs. Attached Figure Description

[0022] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Figure 1 This is a flowchart of the ultraviolet-visible image registration method in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the combination of basic registration relationships and correction relationships in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the multi-distance registration parameter library call in an embodiment of the present invention; Figure 4 This is a diagram showing the registration results in this embodiment of the invention and the registration results of the comparison method. Detailed Implementation

[0023] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and, together with the embodiments of the present invention, serve to illustrate the principles of the present invention.

[0024] Example 1 One embodiment of the present invention discloses a method for ultraviolet-visible image registration; such as Figure 1 As shown, it includes: S1. Establish a dual-state imaging system, fix the relative installation positions of the ultraviolet detector component and the visible light detector component, and set the first imaging state and the second imaging state of the ultraviolet detector; the first imaging state is the state without loading a solar-blind ultraviolet filter, and the second imaging state is the state with loading a solar-blind ultraviolet filter. S2. In the first imaging state, based on the acquired ultraviolet image and visible light image, establish a basic registration relationship from ultraviolet image to visible light image; S3. While keeping the relative positions of the two detectors unchanged, switch to the second imaging state. Based on the corresponding control points between the solar-blind ultraviolet image in the second imaging state and the ultraviolet image in the first imaging state, establish a state correction relationship from the second imaging state to the first imaging state. S4. Combine the basic registration relationship with the state correction relationship to obtain the final registration relationship from the solar-blind ultraviolet image to the visible light image in the second imaging state.

[0025] Specifically, in the dual-state imaging system established by S1, the ultraviolet detector assembly and the visible light detector assembly are fixedly mounted on the same rigid bracket or shared platform to ensure that the two maintain their relative poses unchanged throughout the calibration and detection process, and that the two detectors have a common field of view.

[0026] The first imaging state of the system is the filterless state where the ultraviolet detector is not loaded with a solar-blind ultraviolet filter. In this state, the ultraviolet detector has a large light transmittance and a high image signal-to-noise ratio, which can be used for high-precision calibration. The second imaging state of the system is the actual detection state after the ultraviolet detector has a solar-blind ultraviolet narrowband filter loaded at the front of the lens. This state corresponds to the actual working conditions of the system.

[0027] S2 includes two sub-processes: single-detector imaging parameter calibration and dual-detector image registration. S2-1, Single detector imaging parameter calibration; Obtain the intrinsic parameters, distortion parameters or equivalent imaging correction parameters of the ultraviolet detector assembly and the visible light detector assembly respectively; This step uses a high-precision checkerboard calibration target to calibrate imaging parameters in the first imaging state.

[0028] During calibration, a high-precision checkerboard calibration target is placed within the common field of view of the ultraviolet detector assembly and the visible light detector assembly, and several ultraviolet calibration images and visible light calibration images under different postures are acquired respectively. The corner detection algorithm was used to extract the coordinates of the checkerboard corners in each image, and the intrinsic parameter matrix of the ultraviolet detector was solved using the Zhang Zhengyou calibration method. Distortion coefficient vector and the intrinsic parameter matrix of the visible light detector. Distortion coefficient vector ; The images are processed to remove distortion based on their respective distortion parameters, resulting in standardized ultraviolet and visible light standard images.

[0029] S2-2, Dual detector image registration: After completing distortion correction and imaging standardization using correction parameters, establish the basic registration relationship between the ultraviolet standard image coordinate system and the visible light standard image coordinate system in the first imaging state; Specifically, the basic registration relation is expressed as: ; in, These are the homogeneous coordinates of the pixels in the ultraviolet image under the first imaging state. These are the pixel coordinates of the ultraviolet image; These are the homogeneous coordinates of the corresponding pixel in the visible light image under the first imaging state. These are the pixel coordinates of the visible light image; Based on the registration relationship, the transformation matrix describes the mapping between the image coordinate systems of the two detectors.

[0030] Preferably, The homography matrix is ​​used for characterization. The homography matrix is ​​a 3×3 matrix (with 8 degrees of freedom in a homogeneous sense), which can fully describe the perspective projection relationship between the two detectors. Its specific form is as follows: ; The elements of the basic registration relationship homography matrix; Basic registration relationship The process of determining includes: On the distortion-corrected standard image, select at least four pairs of corresponding points from uniformly distributed ultraviolet and visible light images, and solve the initial homography matrix using the Direct Linear Transformation (DLT) method. The initial homography matrix satisfy ; With the goal of minimizing reprojection error, the matrix parameters are refined using the LM (Levenberg-Marquardt) nonlinear optimization method. To improve the robustness of outlier points such as those mis-extracted from corner points, the Random Sampling Consensus Algorithm (RANSAC) can be used to filter interior points of corresponding point pairs before solving the problem. This process removes mismatched point pairs and yields a high-precision basic registration relationship. .

[0031] The basic registration relationship is not equivalent to the imaging parameters of a single detector, but rather describes the spatial correspondence between the output images of two detectors.

[0032] Specifically, in S3, the state correction relationship between the solar-blind ultraviolet image coordinate system in the second imaging state and the ultraviolet image coordinate system in the first imaging state is a coordinate mapping relationship. The equivalent focal length change, principal point offset, residual distortion, image rotation, scale change and local coordinate deviation caused by the filter loading are uniformly represented as two-dimensional coordinate transformation, and the mapping between the ultraviolet image coordinate system in the second imaging state and the first imaging state is directly established.

[0033] Includes the following steps: S3-1, Target Placement and Image Acquisition; While keeping the relative orientation of the ultraviolet detector assembly and the visible light detector assembly unchanged, the solar-blind ultraviolet filter is loaded onto the front end of the lens of the ultraviolet detector assembly, so that the system enters the second imaging state.

[0034] An LED dot matrix target composed of several ultraviolet LED point light sources arranged in a known geometric pattern is used as the control point carrier; the central radiation wavelength of the ultraviolet LED point light source is located within the passband of the solar-blind ultraviolet filter or the effective response band of the solar-blind ultraviolet detector assembly. In the first imaging state, the LED dot matrix target is placed within the field of view of the ultraviolet detector assembly, and an ultraviolet target image in the first state is acquired. The coordinate set of each emitting point in the ultraviolet image in the filterless state is obtained by the centroid extraction algorithm. ; Keeping the LED dot matrix target position unchanged, switch to the second imaging state, acquire the ultraviolet target image in the second state, and obtain the coordinate set of each emitting point in the ultraviolet image in the state with the filter loaded by the centroid extraction algorithm. ; The LED dot matrix target also includes visible light identifiable markers to assist in target attitude maintenance, control point number identification, or manual verification.

[0035] S3-2, Control point selection and correspondence establishment; Match the centroid coordinates of the luminescent points in the ultraviolet target image in the first state and the ultraviolet target image in the second state to establish a set of corresponding control point pairs. .

[0036] The number of effective control points There should be no fewer than three groups, and each control point should be non-collinear to ensure the uniqueness and numerical stability of the state correction relationship solution.

[0037] S3-3, Solving for state correction relationships; Let the first image in the ultraviolet image under the second imaging state be... The coordinates of the control points are: Its corresponding coordinates in the ultraviolet image under the first imaging state are .

[0038] The state correction relationship The affine transformation matrix is ​​characterized by an affine transformation model. For a 2×3 matrix, satisfying:

[0039] in, These are elements of the linear transformation submatrix, comprehensively describing the rotation, scaling, and shearing components; The translation component describes the overall image offset.

[0040] use A system of linear equations is constructed using pairs of non-collinear corresponding control points, with the objective of minimizing the reprojection error: ; in, These are the homogeneous coordinates of the control point in the second imaging state.

[0041] The affine transformation matrix is ​​solved directly using the least squares method. When outlier noise exists at the control points, a random sampling consensus algorithm is used for robust estimation to further improve the robustness of the state correction relationship.

[0042] The state correction relationship Instead of requiring the complete imaging parameters of the ultraviolet detector assembly to be re-obtained in the second imaging state as a necessary condition, the equivalent focal length change, principal point offset, residual distortion, image rotation, scale change or local coordinate deviation that may be caused by the loading of the filter are uniformly characterized as the coordinate transformation.

[0043] Specifically, in S4, the basic registration relationship is combined with the state correction relationship to obtain the final registration relationship applicable to the second imaging state, i.e., the actual detection condition with the solar blind filter loaded.

[0044] Includes the following steps: S4-1, Composition of transformation matrices; The basic registration relationship Relationship with the state correction Perform sequential compounding to obtain the final registration relationship. : ; in, For mapping combination; The composite operation means that for any pixel in the ultraviolet image under the second imaging state... First, the state correction relationship is established. Equivalent coordinates mapped to the first imaging state Then, based on the aforementioned basic registration relationship Mapped to the corresponding coordinates in the visible light image coordinate system .

[0045] Right now .

[0046] S4-2, Explicit expression in matrix form; Expanding the above equation in matrix form, let... Represented by a homography matrix (3×3), With affine transformation matrix ( Homogeneous expansion to If ), then the above formula can be written as: ; in, This is a normalization factor to ensure that the output is pixel coordinates in homogeneous coordinates.

[0047] S4-3. Verification and application of registration results; The final registration relationship The image is applied to the solar-blind ultraviolet image acquired in the second imaging state to achieve pixel-level spatial registration with the visible light image.

[0048] The final registration relationship is directly applicable to actual solar-blind ultraviolet detection conditions, without the need for complete recalibration in the second imaging state.

[0049] The registration accuracy of the final registration relationship can be comprehensively evaluated using point-level geometric error and task area position error.

[0050] To evaluate the effectiveness of the final registration relationship, this embodiment uses the average reprojection error of the verification control points, the root mean square reprojection error, and the centroid shift of the region of interest in the discharge for comprehensive evaluation.

[0051] For the first [unrelated] not involved in solving the current mapping relationship For each verification control point, its ultraviolet coordinates in the second imaging state are first registered using the final registration relationship. Map the coordinates to the visible light image coordinate system and compare them with the reference coordinates of the control point in the visible light image to obtain the corresponding reprojection error. Calculate the average reprojection error and root mean square reprojection error (RMSE) for all verification control points.

[0052] For actual discharge detection images, the bright discharge spot region is first extracted from the ultraviolet image in the second imaging state to obtain the region of interest (ROI). Then, the ROI is mapped to the visible light image coordinate system, and the centroid position of the mapped region is calculated. This centroid position is compared with a manually verified reference position, an independent reference calibration position, or a known discharge structure position to obtain the centroid offset of the ROI.

[0053] Among them, the average reprojection error and the root mean square reprojection error (RMSE) are used to evaluate the overall geometric registration accuracy, and the centroid shift of the region of interest in discharge is used to evaluate the positional consistency of the ultraviolet discharge response in the actual visible light device structure.

[0054] The diagram illustrating the generation and combination of the basic registration relations and correction relations from S2 to S4 is shown below. Figure 2 As shown.

[0055] Preferably, the method in this embodiment further includes establishing a multi-distance registration parameter library; The process involves establishing a multi-distance registration parameter library, and then performing the steps of obtaining the basic registration relationship, the state correction relationship, and the combined registration relationship at multiple representative detection distances to obtain the basic registration relationship, the state correction relationship, and the final registration relationship corresponding to each distance. During actual detection, the system searches or interpolates the data in the multi-distance registration parameter library based on the current detection distance.

[0056] Furthermore, the interpolation call includes: if the current detection distance matches a certain calibrated distance, the corresponding final registration relationship is directly called; if the current detection distance is between two calibrated distances, the parameters of the basic registration relationship and the state correction relationship are interpolated separately and then combined to obtain the final registration relationship under the current detection distance.

[0057] Specifically, multi-distance registration parameter library calls, such as Figure 3 As shown, since the image registration relationship changes with the detection distance and lens focus state, after completing the calibration at a single distance, the registration relationship is further refined at multiple representative detection distances. Repeat steps S2 to S4 to obtain the basic registration relationship corresponding to each distance. State correction relationship and final registration relationship Construct a multi-distance registration parameter library: ; in: ; Furthermore, during actual detection, based on the input current detection distance... (This can be obtained from sensor detection, target size estimation, or manual input), and can be searched and retrieved from the parameter library: If With a certain calibrated distance If they match, then call directly. ;like Located at two calibrated distances and Between them, the preferred method is to compare the basic registration relationship. Relationship with state correction Perform parameter interpolation to obtain and Recombining them, we get: ; Ultimately, Applying a load filter to the solar-blind ultraviolet image of the current frame to achieve automatic registration with the visible light image: .

[0058] In summary, the UV-Vis image registration method of this embodiment does not perform complete UV-Vis calibration directly under the condition of loading a solar-blind UV filter. Instead, it first obtains a stable basic registration relationship using the unfiltered state, and then obtains a state correction relationship using the filter-loaded state. This avoids the instability problem caused by direct complete calibration in low signal-to-noise ratio UV images after loading a solar-blind UV filter. The method compensates for possible changes in optical path, image shift, scale change, or local residuals caused by loading a solar-blind ultraviolet filter through state correction relations, so that the basic registration relation obtained in the filterless state can be applied to the actual detection conditions after loading the filter.

[0059] The method decouples and stores the "basic registration relationship" and the "state correction relationship", which can establish a registration parameter library at different detection distances or different focusing states, thereby improving the automation and engineering adaptability of UV-Vis registration in multi-distance scenarios.

[0060] The state correction process in this method only requires a small number of corresponding control points. Compared with recalibrating under the condition of loading a solar-blind ultraviolet filter, it has the advantages of simplified operation, convenient parameter updates and low engineering deployment costs.

[0061] In this embodiment, an imaging distance of 50 cm was selected to verify the registration effect. The ultraviolet channel acquired discharge response images under the same 7.5V discharge condition, while the visible light channel simultaneously acquired images of the electrodes and surrounding structures. The comparison method employed a checkerboard target direct calibration to establish the registration relationship between the ultraviolet and visible light images. Both methods were applied to the same ultraviolet discharge image to eliminate the influence of changes in discharge state on the comparison results.

[0062] like Figure 4 As shown, the pink outline represents the region of interest mapping result obtained using the final registration relationship of this invention, and the red outline represents the mapping result obtained using the direct calibration method with an open checkerboard target. Figure 4 As can be seen, after adopting the method of the present invention, the position of the electrode gap structure in the ultraviolet discharge region and the visible light image has a higher consistency, with an average reprojection error of 3.59 px, a root mean square reprojection error of 3.70 px, and an average centroid offset of 2.788 px. In contrast, the mapping contour obtained by the comparison method has a more obvious positional offset, with an average reprojection error of 14.15 px, a root mean square reprojection error of 14.29 px, and an average centroid offset of 15.262 px.

[0063] The above results indicate that by combining the basic registration relationship in the first imaging state with the state correction relationship in the second imaging state, the registration residual caused by optical path changes, coordinate offsets and local imaging differences in the filter loading state can be reduced, thereby improving the positional accuracy of the solar-blind ultraviolet discharge region mapped to the visible light structure image.

[0064] Example 2 This embodiment discloses an ultraviolet-visible image registration device, comprising: Rigid installation platform; An ultraviolet detector assembly, fixed on the rigid mounting platform, includes a solar-blind ultraviolet filter and a filter switching mechanism, wherein the filter switching mechanism is configured to switch between a first imaging state without the solar-blind ultraviolet filter and a second imaging state with the solar-blind ultraviolet filter loaded. The visible light detector assembly is fixed on the rigid mounting platform, maintaining a constant relative orientation with the ultraviolet detector assembly; The processor, which is communicatively connected to the ultraviolet detector assembly and the visible light detector assembly, is configured to: In the first imaging state, a basic registration relationship from ultraviolet image to visible light image is established based on the acquired ultraviolet image and visible light image; While keeping the relative positions of the two detectors unchanged, switch to the second imaging state. Based on the corresponding control points between the solar-blind ultraviolet image in the second imaging state and the ultraviolet image in the first imaging state, establish a state correction relationship from the second imaging state to the first imaging state. The state correction relationship uniformly represents the equivalent focal length change, principal point offset, residual distortion, image rotation, scale change and local coordinate deviation caused by filter loading as a two-dimensional coordinate transformation. The basic registration relationship is combined with the state correction relationship to obtain the final registration relationship from the solar-blind ultraviolet image to the visible light image in the second imaging state.

[0065] In the first imaging state, a basic registration relationship from ultraviolet image to visible light image is established based on the acquired ultraviolet image and visible light image; The basic registration relationship is combined with the state correction relationship to obtain the final registration relationship from the solar-blind ultraviolet image to the visible light image in the second imaging state.

[0066] Furthermore, it also includes target components. The target assembly includes a checkerboard target and an LED dot matrix target. The checkerboard calibration target is used for calibrating imaging parameters and establishing basic registration relationships in the first imaging state. The LED dot matrix target is composed of several ultraviolet LED point light sources arranged in a known geometric pattern, and is used to establish the state correction relationship in the second imaging state.

[0067] Furthermore, the processor is also configured to: A multi-distance registration parameter library is established. The establishment of the basic registration relationship, the establishment of the state correction relationship, and the combination of the final registration relationship are performed at multiple representative detection distances to obtain the basic registration relationship, state correction relationship, and final registration relationship corresponding to each distance. The current detection distance is used to search or interpolate the multi-distance registration parameter library.

[0068] The more specific technical details and beneficial effects in this embodiment are the same as in Embodiment 1. Please refer to them for details, and they will not be repeated here.

[0069] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for ultraviolet-visible image registration, characterized in that, include: A dual-state imaging system was established, with the relative installation positions of the ultraviolet detector assembly and the visible light detector assembly fixed, and the first and second imaging states of the ultraviolet detectors were set. The first imaging state is the state without a solar-blind ultraviolet filter, and the second imaging state is the state with a solar-blind ultraviolet filter. In the first imaging state, a basic registration relationship from ultraviolet image to visible light image is established based on the acquired ultraviolet image and visible light image; While keeping the relative positions of the two detectors unchanged, switch to the second imaging state. Based on the corresponding control points between the solar-blind ultraviolet image in the second imaging state and the ultraviolet image in the first imaging state, establish a state correction relationship from the second imaging state to the first imaging state. The basic registration relationship is combined with the state correction relationship to obtain the final registration relationship from the solar-blind ultraviolet image to the visible light image in the second imaging state.

2. The ultraviolet-visible image registration method according to claim 1, characterized in that, The state correction relationship is a coordinate mapping relationship, which uniformly represents the equivalent focal length change, principal point offset, residual distortion, image rotation, scale change and local coordinate deviation caused by the filter loading as a two-dimensional coordinate transformation, and directly establishes the mapping between the ultraviolet image coordinate system of the second imaging state and the first imaging state.

3. The ultraviolet-visible image registration method according to claim 1 or 2, characterized in that, The state correction relationship is represented by an affine transformation model, and the affine transformation matrix is ​​a 2×3 matrix. Its linear transformation submatrix Used to comprehensively describe rotation, scaling, shearing components, and translation components. Used to describe the overall image offset.

4. The ultraviolet-visible image registration method according to claim 3, characterized in that, The number of corresponding control points is no less than three groups, and each control point is not collinear with the others; the state correction relationship is solved directly by the least squares method to obtain the affine transformation matrix parameters, or when there is outlier noise in the control points, the random sampling consensus algorithm is used for robust estimation.

5. The ultraviolet-visible image registration method according to claim 1 or 2, characterized in that, In the acquisition state correction relationship, an LED dot matrix target composed of several ultraviolet LED point light sources arranged in a known geometric pattern is used as the control point carrier; the central radiation wavelength of the ultraviolet LED point light source is located within the passband of the solar-blind ultraviolet filter or the effective response band of the solar-blind ultraviolet detector assembly; in the first imaging state and the second imaging state, the centroid of the same group of LED light-emitting points is extracted to obtain the corresponding control point coordinates, and the target position and the relative pose of the two detectors remain unchanged during the acquisition process.

6. The ultraviolet-visible image registration method according to claim 1, characterized in that, Also includes: A multi-distance registration parameter library is established. The steps of obtaining the basic registration relationship, obtaining the state correction relationship, and combining the registration relationship are executed at multiple representative detection distances to obtain the basic registration relationship, state correction relationship, and final registration relationship corresponding to each distance. During actual testing, the current testing distance is used to search or interpolate from the multi-distance registration parameter library.

7. The ultraviolet-visible image registration method according to claim 6, characterized in that, The interpolation call includes: if the current detection distance matches a certain calibrated distance, the corresponding final registration relationship is directly called; if the current detection distance is between two calibrated distances, the parameters of the basic registration relationship and the state correction relationship are interpolated separately and then combined to obtain the final registration relationship under the current detection distance.

8. A UV-Vis image registration system, characterized in that, include: Rigid installation platform; An ultraviolet detector assembly, fixed on the rigid mounting platform, includes a solar-blind ultraviolet filter and a filter switching mechanism, wherein the filter switching mechanism is configured to switch between a first imaging state without the solar-blind ultraviolet filter and a second imaging state with the solar-blind ultraviolet filter loaded. The visible light detector assembly is fixed on the rigid mounting platform, maintaining a constant relative orientation with the ultraviolet detector assembly; The processor, which is communicatively connected to the ultraviolet detector assembly and the visible light detector assembly, is configured to: In the first imaging state, a basic registration relationship from ultraviolet image to visible light image is established based on the acquired ultraviolet image and visible light image; While keeping the relative positions of the two detectors unchanged, switch to the second imaging state. Based on the corresponding control points between the solar-blind ultraviolet image in the second imaging state and the ultraviolet image in the first imaging state, establish a state correction relationship from the second imaging state to the first imaging state. The state correction relationship uniformly represents the equivalent focal length change, principal point offset, residual distortion, image rotation, scale change and local coordinate deviation caused by filter loading as a two-dimensional coordinate transformation. The basic registration relationship is combined with the state correction relationship to obtain the final registration relationship from the solar-blind ultraviolet image to the visible light image in the second imaging state. In the first imaging state, a basic registration relationship from ultraviolet image to visible light image is established based on the acquired ultraviolet image and visible light image; The basic registration relationship is combined with the state correction relationship to obtain the final registration relationship from the solar-blind ultraviolet image to the visible light image in the second imaging state.

9. The ultraviolet-visible image registration system according to claim 8, characterized in that, It also includes target components. The target assembly includes a checkerboard target and an LED dot matrix target. The checkerboard calibration target is used for calibrating imaging parameters and establishing basic registration relationships in the first imaging state. The LED dot matrix target is composed of several ultraviolet LED point light sources arranged in a known geometric pattern, and is used to establish the state correction relationship in the second imaging state.

10. The ultraviolet-visible image registration system according to claim 8, characterized in that, The processor is also configured to: A multi-distance registration parameter library is established. The establishment of the basic registration relationship, the establishment of the state correction relationship, and the combination of the final registration relationship are performed at multiple representative detection distances to obtain the basic registration relationship, state correction relationship, and final registration relationship corresponding to each distance. The current detection distance is used to search or interpolate the multi-distance registration parameter library.

Images