Correction method and system for industrial close-range photogrammetry image

Abstract

The invention discloses a correction method and system for an industrial close-range photogrammetry image, and relates to the technical field of computer vision, and the method comprises the steps: S100, constructing a cross-time-scale light path observation baseline under a multi-time-scale sampling condition, carrying out the combined extraction operation of brightness distribution and phase gradient on each frame of image, generating an anchor point sequence, and carrying out the detection of the anchor point sequence; an interference risk map is constructed according to the change trend of the optical path difference and is used as a dynamic constraint basis for feature recognition; and S200, performing phase polarization consistency auditing based on the anchor point sequence and the interference risk map, performing pixel-by-pixel tracking for a phase mutation region in the brightness field, extracting an interference pseudo peak with a phase inversion characteristic, and outputting a pseudo edge candidate data set. According to the method, a time anchor point and an interference map are constructed, and interference sensing and early warning are realized; a real edge is extracted in combination with phase auditing and optical flow registration, imaging geometry is reconstructed, distortion is optimized, interference is suppressed in combination with phase conjugation and time rearrangement, and closed-loop control and high-precision image correction are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision, in particular to a correction method and system for industrial close-range photogrammetry images. BACKGROUND

[0002] The image correction for industrial close-range photogrammetry refers to the whole process of implementing systematic image correction processing on images obtained by close-range shooting in the process of industrial measurement or three-dimensional reconstruction, so as to eliminate geometric and optical errors. Due to the complex lighting, space limitation, mechanical vibration, high reflective metal surface and other conditions of the camera in the industrial environment, the original image is prone to distortion, tilt, scale distortion, focal length drift, uneven lighting and other problems in the imaging process, thereby significantly reducing the measurement accuracy. The image correction process establishes a high-precision camera imaging model, parameterizes modeling and inverse calculation of lens distortion, principal point offset, pixel nonlinear deformation, and restores the true geometric relationship of the image by combining the spatial reference point or calibration board information, so that the image coordinates and the actual physical space coordinates are accurately corresponding. At the same time, the image correction process introduces a lighting balance and color consistency correction mechanism to compensate and reconstruct the shadow, reflection and uneven brightness area, so as to ensure that the corrected image can truly reflect the structure, shape and size characteristics of the measured object, and provide high-reliability visual data support for industrial detection, assembly precision control and digital modeling.

[0003] In the prior art, when the measured object has a high-reflectivity metal surface, the incident light will undergo multiple reflections and superpositions on the surface, forming a complex optical path interference effect. Due to the small changes in reflection angle and surface roughness, the phase superposition caused by the optical path difference will cause the phase inversion phenomenon in the local pixel area, thereby forming a false peak signal with abnormal mutation of brightness in the imaging result. The false peak presents similar characteristics to the true physical boundary in the gray gradient, so that the edge detection algorithm based on brightness change in the prior art is easy to misidentify it as an actual edge line, thereby introducing false feature points in the feature point extraction stage. With the accumulation of mis-matching points, the spatial registration model is geometrically unstable, which causes the grid topological structure to be broken or distorted in three-dimensional reconstruction, seriously affecting the spatial accuracy and geometric continuity of the close-range photogrammetry result.

[0004] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a correction method and system for industrial close-range photogrammetry images to solve the problems in the background.

[0006] In order to achieve the above object, the present application provides the following technical scheme: a correction method for industrial close-range photogrammetry images, comprising the following steps: S100, under multi-time sampling conditions, constructing a cross-time baseline of light path observation, performing joint extraction of brightness distribution and phase gradient on each image frame, generating an anchor point sequence with time continuity, and constructing an interference risk map according to the light path difference variation trend, which is used as a dynamic constraint basis for feature recognition; S200, performing phase polarization consistency audit based on the anchor point sequence and the interference risk map, performing pixel-by-pixel tracking on the phase mutation area in the brightness field, extracting interference false peaks with phase inversion characteristics, and outputting a false edge candidate dataset containing spatial position and phase state; S300, performing multi-view optical flow depth coupling registration based on the false edge candidate dataset and the phase audit result, removing edge points with time drift through cross-frame optical flow consistency detection, and converting edge regions with stable morphological characteristics into a real edge evidence dataset; S400, reconstructing the imaging geometry based on the real edge evidence dataset, performing real-time adjustment of the camera internal and external parameters, and performing online inversion and recursive optimization of the nonlinear distortion parameters, establishing a self-consistent correction field covering the whole time sequence, and realizing the alignment of the light path system and the imaging coordinate system; S500, performing dynamic regulation operation based on the self-consistent correction field, performing periodic reverse modulation on the interference false peak energy distribution through phase conjugate polarization rotation scanning combined with reversible time grid rearrangement mechanism, continuously suppressing the phase inversion area, and constructing a closed-loop control process for interference detection and interference source suppression.

[0007] Preferably, step S100 comprises: establishing an imaging structure with a spatial calibration reference under multi-time sampling conditions, and acquiring an image frame sequence containing brightness information and phase influence through a multi-spectral imaging device with adjustable focal length; Based on the acquired image frame sequence, joint feature data of brightness distribution and phase gradient are extracted, and the pixels are analyzed for local gray scale continuity and phase variation trend, and a joint feature point dataset is constructed; Based on the constructed joint feature point dataset, the regions with consistent variation characteristics in the time sequence are selected as anchor point candidate regions, and spatial back projection verification is performed combined with the light path observation baseline to determine the time anchor point sequence; Based on the time anchor point sequence and the light path variation trend, the gray scale fluctuation and phase oscillation analysis is performed on the non-anchor point region, and an interference risk map for dynamic constraint is constructed and the interference risk region distribution existing in the image is recorded.

[0008] Preferably, step S200 comprises: Screening the continuous pixel area with high interference risk score as the audit area in the image pixel level range based on the time anchor point sequence and the interference risk atlas, and establishing the spatial proximity relationship with the anchor point; Performing comparison analysis of the brightness change curve and the anchor point brightness trajectory for each pixel point in the audit area, and identifying the phase inconsistent area in combination with the gray response difference under different polarization directions; Performing pixel-by-pixel time sequence tracking on each pixel point in the phase inconsistent area, extracting the brightness polarity reversal point and the polarization response synchronization feature, and determining whether there is a phase reversal phenomenon; According to the determination result, output the pseudo-edge candidate data set containing the spatial position and phase state information, perform boundary fitting and cross verification with the time anchor point, and form the pseudo-peak trajectory set with clear structure.

[0009] Preferably, in the pixel-by-pixel time sequence tracking process, the phase reversal phenomenon is jointly determined according to the distribution frequency of the brightness polarity reversal point, the spatial jump amplitude and the polarization response change synchronization, and only when the three conditions are met at the same time, the corresponding pixel is marked as an interference pseudo-peak.

[0010] Preferably, step S300 comprises: Based on the spatial position index and the phase history information of the pseudo-edge candidate data set, the corresponding area in the multi-view image sequence is retrieved, and the observation frame sequence with time sequence relationship is constructed for subsequent analysis; Performing optical flow field frame-by-frame tracking on the pseudo-edge area in each observation frame sequence, establishing the pixel motion trajectory between the continuous image frames, and introducing the parallax to calculate the spatial depth consistency of each pixel; Based on the optical flow tracking result, performing cross-frame consistency detection on all pseudo-edge points, combining the phase response trajectory and the interference score change, and eliminating the non-stable edge points with time sequence drift and spatial jump; Constructing all the edge regions that pass the consistency verification into the real edge evidence data set, extracting the spatial position, depth estimate and contour shape information, and providing high reliability input for subsequent geometric reconstruction.

[0011] Preferably, in the process of constructing the real edge evidence data set, the edge region that passes the consistency verification is subjected to sub-pixel level boundary fitting, and an edge shape template is generated, which is used as an edge constraint input in subsequent imaging geometric structure reconstruction.

[0012] Preferably, step S400 comprises: Based on the edge coordinates, optical flow trajectory, phase feature and depth value in the real edge evidence data set, performing spatial back projection operation on each frame of image, and establishing the projection geometric relationship of the imaging light path; An error distribution map is constructed based on the projection geometry, and dynamic adjustment operations are performed on the camera's intrinsic and extrinsic parameters to optimize the camera's imaging center position, focal length, optical axis offset, and external attitude information. After adjusting the camera parameters, nonlinear distortion parameters are modeled and recursively optimized based on the theoretical trajectory and actual position of the edge points to compensate for radial and tangential distortion errors. The camera parameters of each frame are fused with the distortion model to construct a self-consistent correction field covering the entire time series, which is used to guide the geometric alignment and optical path consistency correction of all image frames in space.

[0013] Preferably, in the recursive optimization process of nonlinear distortion parameters, the distortion model is locally reconstructed based on the matching error of edge contours in each frame image, and the parameter weights are dynamically adjusted using a spatial consistency evaluation mechanism to improve the geometric correction accuracy.

[0014] Preferably, step S500 includes: Based on the imaging parameters, optical path structure and distortion correction model recorded in the self-consistent correction field, an optical path control structure with temporal continuity and spatial consistency is constructed, and a target control node with phase reversal characteristics is identified. By using a phase conjugate polarization rotation scanning mechanism, the polarization state of the incident light is adjusted within the high-reflection region to form a structure that is phase-complementary to the reflection path, thereby interfering with the phase superposition process of the interference pseudo-peaks. By reversible temporal raster rearrangement mechanism, the evolution trajectory of interference intensity in the image frame sequence is reconstructed, the inter-frame temporal structure is adjusted to weaken the accumulation of pseudo-peak energy, and a temporal dimension interference control strategy is constructed. Based on the aforementioned dual phase and time adjustment method, a closed-loop control process for interference detection and interference source suppression is constructed. The control parameters are set before image acquisition and the control command for the next frame is updated based on the response feedback after acquisition.

[0015] The present invention also provides a correction system for industrial close-range photogrammetry images, comprising: The time-scaled optical path observation module is configured to construct a cross-time-scale optical path observation baseline under multi-time-scale sampling conditions, perform joint extraction operations of brightness distribution and phase gradient on each frame image, generate a time-continuous anchor point sequence, and construct an interference risk map based on the trend of optical path difference change, which is used as the dynamic constraint basis for feature recognition. The phase consistency audit module is configured to perform phase polarization consistency audit based on anchor point sequence and interferometric risk map. It performs pixel-by-pixel tracking of phase abrupt change regions in the brightness field, extracts interferometric pseudo-peaks with phase reversal characteristics, and outputs a pseudo-edge candidate dataset containing spatial location and phase state. The optical flow coupling registration module is configured to perform multi-view optical flow deep coupling registration based on the pseudo edge candidate dataset and phase audit results. It eliminates edge points with temporal drift through cross-frame optical flow consistency detection and transforms edge regions with stable morphological features into real edge evidence datasets. The geometric correction and reconstruction module is configured to reconstruct the imaging geometry based on the real edge evidence dataset, perform real-time adjustment of camera intrinsic and extrinsic parameters, perform online inversion and recursive optimization of nonlinear distortion parameters, establish a self-consistent correction field covering the entire time series, and achieve alignment between the optical path system and the imaging coordinate system. The interference suppression control module is configured to perform dynamic control operations based on the self-consistent correction field. Through phase conjugate polarization rotation scanning combined with a reversible time grid rearrangement mechanism, it performs periodic reverse modulation on the energy distribution of interference pseudo-peaks, continuously suppresses the phase reversal region, and constructs a closed-loop control process for interference detection and interference source suppression.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention achieves dynamic perception and risk warning of image interference potential by constructing a temporally continuous anchor point sequence and interference risk map. Combined with phase polarization consistency auditing and pixel-by-pixel tracking, it accurately identifies and isolates pseudo-edge regions with phase reversal characteristics. Under a multi-view optical flow registration mechanism, it further eliminates unstable feature points disturbed by interference, thereby extracting highly reliable true edge evidence. Imaging geometry reconstruction and distortion optimization based on this evidence dataset enable precise full-time correction of imaging parameters, establishing an integrated alignment relationship between the optical path and the image. Furthermore, by employing phase conjugate polarization scanning and temporal grid rearrangement strategies, an active control and periodic suppression mechanism for interference energy is formed, constructing a closed-loop control process from interference detection to adaptive control, achieving full-process suppression and feedback adjustment of the interference source. The overall scheme possesses the synergistic advantages of dynamic perception, precise correction, and adaptive suppression, ultimately achieving accurate identification and stable elimination of interference pseudo-peaks in industrial close-up images, providing a solid data foundation and robust image assurance for high-precision 3D reconstruction and precision visual measurement. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a flowchart of the method for correcting industrial close-range photogrammetric images according to the present invention.

[0019] Figure 2 This is a schematic diagram of the module of the correction system for industrial close-range photogrammetric images of the present invention. Detailed Implementation

[0020] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0021] This invention provides, for example Figure 1 The calibration method shown for industrial close-range photogrammetry images includes the following steps: S100 constructs a cross-timescale optical path observation baseline under multi-timescale sampling conditions, performs joint extraction operations of brightness distribution and phase gradient on each frame of image, generates a time-continuous anchor point sequence, and constructs an interference risk map based on the trend of optical path difference change, which is used as the dynamic constraint basis for feature recognition. In industrial close-range photogrammetry image processing, to address the interference spurious peak problem caused by highly reflective metal surfaces, a technical approach is proposed to construct a cross-timescale optical path observation baseline under multi-timescale sampling conditions. By jointly extracting the image brightness distribution and phase gradient, a temporally continuous anchor point sequence is generated. Furthermore, an interference risk map is constructed based on the changing trend of the optical path difference, providing a dynamic constraint basis for subsequent feature recognition and spurious feature removal. The specific implementation steps are as follows: In an industrial measurement environment with a stable sampling frequency and a fixed spatial reference, an adjustable-focus multispectral imaging device is deployed, along with a spatial calibration structure capable of sub-pixel-level positioning. This structure consists of high-precision calibration points distributed at different locations in three-dimensional space, used for imaging benchmark comparison at multiple time points. By acquiring images of the same static target at consecutive time points, a set of time-series image frames containing brightness information and phase interference effects is obtained. The sampling time interval is dynamically adjusted according to external interference environments to ensure comparability between image sequences even under conditions of changes in light source stability, small vibration disturbances, or changes in surface reflection. By extracting the pixel distribution changes of the calibration structure in each frame, a preliminary cross-timescale optical path observation baseline is constructed, recording the optical path information from incident to imaging at different times, and using this as a timescale reference for subsequent feature stability.

[0022] Based on the established cross-timescale optical path observation baseline, brightness distribution extraction is performed on the regions containing target objects in each frame of the image. This extraction employs a pixel-level grayscale layering approach, scanning the local neighborhood of each pixel to capture the continuity and stability of grayscale changes. Simultaneously, phase gradient extraction is performed on the image, analyzing the local phase tilt characteristics after light interference on the target surface. This step derives the spatial gradient of the phase propagation path by calculating the direction of grayscale trend changes for each pixel across different time frames. Brightness distribution reflects the intensity variation of surface reflected energy density, while phase gradient reflects the trend of optical path changes caused by surface morphology perturbations. By superimposing time-series data of brightness and phase at the same pixel location and comparing their spatial variation patterns, a joint feature point dataset describing the continuous changes in the image is formed, providing multi-dimensional support for constructing stable anchor points.

[0023] Based on the joint feature extraction of brightness and phase, pixel regions that exist and exhibit consistent trends across multiple consecutive frames are selected as candidate anchor points with temporal continuity, adhering to the dual criteria of temporal consistency and spatial stability. To ensure the stability of the anchor points, spatial geometric constraints are further introduced. Each candidate anchor point is back-projected onto a three-dimensional calibration reference frame based on the previously established optical path observation baseline, and the amplitude and directional stability of the corresponding optical path changes at different times are analyzed. Only when a candidate anchor point exhibits smooth brightness changes, consistent phase gradients, and an optical path inversion projection error below a set threshold throughout the entire time series is the region formally defined as a temporal anchor point. The record for each anchor point includes not only its pixel coordinates and grayscale value on the image plane but also its temporal change curve, phase propagation direction, and its geometrically mapped position in physical space, providing precise temporal and spatial labels for subsequent comparison with interferometric modes.

[0024] After the anchor point sequence is established, its ability to track changes in local image regions over time, combined with the optical path difference trend, is used to construct a map reflecting the degree of optical interference risk. Specifically, for each non-anchor point pixel region, its gray-level fluctuation curve in the time series is first retrieved and compared with the nearest anchor point region. If the region exhibits frequent gray-level polarity reversals, high-amplitude gradient jumps, or nonlinear oscillations in the phase change path, it is marked as a high-interference-risk region. Simultaneously, an inter-image interference period matching mechanism is introduced to compare the spatial diffusion pattern of interference intensity and determine whether the interference phenomenon is propagating. Finally, all identified interference regions are quantitatively scored according to dimensions such as optical path difference variation amplitude, brightness perturbation frequency, and phase drift direction, forming an interference risk map covering the entire image region in a two-dimensional matrix. This map not only determines whether there is a risk of false interference peaks in local regions but, more importantly, provides a risk evolution trajectory over time. This provides a dynamic constraint reference for avoiding high-risk regions in subsequent image feature recognition, preventing the inclusion of interference-induced false features in edge extraction or geometric registration processes, thereby improving the spatial reliability of image correction and the consistency of 3D reconstruction.

[0025] S200 performs phase polarization consistency audit based on anchor point sequence and interferometric risk map, performs pixel-by-pixel tracking of phase abrupt change regions in brightness field, extracts interferometric pseudo-peaks with phase reversal characteristics, and outputs a pseudo-edge candidate dataset containing spatial location and phase state. After constructing the anchor point sequence and interference risk map, in order to identify potential interference pseudo-peak regions in the image, a pseudo-edge candidate dataset containing spatial location and phase state is established by performing fine-grained auditing and pixel-by-pixel tracking on regions with significant phase abrupt changes in the brightness field. This provides stable and reliable data support for subsequent feature selection and geometric reconstruction. The specific steps are as follows: Based on the constructed time anchor sequence and corresponding interferometric risk map, region pre-screening is performed on the pixel-level distribution of the image. Specifically, based on the interferometric risk score of each pixel in the interferometric risk map, consecutive pixel regions with risk levels exceeding a set threshold are selected as preliminary audit areas. These regions are typically located in areas with dense specular reflection or rapidly changing surface reflection directions, exhibiting optical anomalies such as grayscale inversion, abnormal brightness amplification, or phase abrupt changes across multiple image frames. Preliminary pixel sets are established for these regions, and their spatial coordinates are associated with the nearest multiple time anchors to form spatial proximity relationships. The key to this step is to leverage the temporal continuity and spatial stability information provided by the anchors to establish a temporal reference benchmark for the fluctuation state of high-risk pixel groups, facilitating subsequent differential analysis during phase analysis.

[0026] Within the selected high-risk pixel region, a phase consistency audit is performed on the brightness sequence of the target pixel region by combining the spatially stable region provided by the anchor point sequence with its temporal behavior curve. Specifically, the brightness change curve of each pixel within the target region is aligned and compared with the brightness change trajectory in the time anchor point. By analyzing whether there is a significant deviation in the fluctuation trend of the gray values ​​of the two within the same time period, it is determined whether the target pixel has experienced phase inconsistency. At the same time, data from the polarization response channel is introduced to statistically analyze the gray value difference of the same pixel under different polarization directions to determine whether there are signs of abrupt changes in optical path at a specific polarization angle. If the pixel exhibits a gray value change direction different from that of the anchor point region in multiple time segments, and shows a periodic enhancement and decay alternating response curve in the polarization channel, then the pixel is determined to be in a phase unstable region, possibly affected by surface interference effects, and is a potential source of false peaks.

[0027] After completing pixel-level phase-polarization consistency auditing, a pixel-by-pixel cascade tracking operation is performed on pixels marked as phase inconsistency regions. This operation uses time as an index, progressively tracing the spatial location, grayscale value, and phase response changes of each anomalous pixel across consecutive image frames to establish a complete temporal evolution trajectory. By analyzing the distribution frequency, spatial jump amplitude, and polarization response synchronicity of brightness polarity reversal points in this trajectory, it is further determined whether a typical phase reversal phenomenon exists, i.e., a brightness pattern resembling a mirror structure appears across multiple frames and alternates with the polarization state. If the above conditions are met, the trajectory is determined to contain interference pseudo-peaks. Simultaneously, a secondary screening is performed based on the pixel's score change trend in the interference risk map to avoid mislabeling non-interference-related brightness anomalies as pseudo-peaks. Finally, key pixel frames are extracted from all pseudo-peak trajectories that have passed consistency auditing and trajectory tracking and are aggregated into a spatially continuous and temporally stable pseudo-peak structure.

[0028] Based on the above judgment results, a pseudo-edge candidate dataset containing spatial location and phase state information is compiled and output. The specific output includes the image frame number of the pseudo-peak pixel, two-dimensional pixel coordinates, historical grayscale change curves, response feature sequence under the polarization channel, phase abrupt change point location, and corresponding interference risk score. In addition, contour reconstruction of the pseudo-edge region is performed, and the trajectories of multiple pseudo-peaks are fitted to the image plane to extract their possible edge morphology features, which are then marked as objects to be removed or strongly constrained regions. During the output process, the pseudo-edge data is cross-validated with the anchor point sequence from the previous stage to ensure that the spatial boundaries of the two types of data do not overlap, avoiding the erroneous removal of valid feature points due to anchor point misjudgment. Finally, a multidimensional dataset containing full-frame spatial coordinates, phase history, polarization response, and interference annotations is formed, providing a well-structured and clearly sourced data foundation for subsequent feature point selection, mismatch removal, and geometric model constraints.

[0029] S300 performs multi-view optical flow deep coupling registration based on pseudo edge candidate dataset and phase audit results. It eliminates edge points with temporal drift through cross-frame optical flow consistency detection and transforms edge regions with stable morphological features into real edge evidence datasets. After extracting the candidate dataset for false edges and auditing the phase polarization consistency, to improve the credibility of edge features in spatial reconstruction, it is necessary to perform deep registration on the false edge regions through multi-view information fusion and optical flow field consistency detection. From this, true edge regions with temporal stability and spatial consistency are selected and constructed as a true edge evidence dataset for subsequent imaging geometry reconstruction and error control. The specific steps are as follows: Based on the spatial location index of the pseudo-edge candidate dataset and the phase history information it carries, these candidate edge regions are synchronously retrieved and feature regions are cropped in multi-angle image sequences. Specifically, for each set of pseudo-edge pixels, the corresponding observation viewpoint image region is extracted from neighboring frames according to its trajectory characteristics in consecutive frames. These regions are then arranged according to a uniform scale, relative viewpoint interval, and temporal order, forming a multi-frame observation sequence with clear temporal relationships and spatial viewpoint distribution characteristics. The purpose of this step is to establish a corresponding region set for each pseudo-edge region in the multi-view images, thereby providing a continuous spatial basis and temporal constraints for subsequent optical flow field coupling analysis. In this process, the polarization response differences recorded in the phase consistency audit results of the previous stage are specifically referenced to pre-screen edge regions that maintain phase characteristic continuity under different viewpoints, prioritizing them as coupling candidate regions.

[0030] For the constructed multi-view image sequence, frame-by-frame optical flow tracing is performed around the pseudo-edge regions to establish a spatial mapping relationship of pixel motion between consecutive frames. Specifically, in each frame, fine-grained motion trend analysis is performed on pixels in the pseudo-edge regions. By comparing the direction and magnitude of grayscale changes at the same location between adjacent frames, the translation, rotation, or scaling trends on the time axis are calculated, forming a continuous motion trajectory. During optical flow tracing, the local structural fidelity of pixels is analyzed, i.e., whether the texture structure around the pixel remains relatively consistent across multiple frames. To enhance the robustness of optical flow tracing, phase abrupt change information extracted in the previous stage is combined to specially mark locations with sudden changes in the trajectory, enabling the identification of potential instability factors in subsequent analysis. Simultaneously, a disparity-matching depth estimation operation is introduced to geometrically reproject the relative positional relationship of each pixel across multiple viewpoints to analyze the actual depth consistency of the pixel in space, thereby eliminating false edge structures that only stand out at a certain angle or time frame but cannot be confirmed by multiple viewpoints.

[0031] After obtaining multi-frame optical flow field tracking results, a cross-frame consistency detection operation is performed on the temporal trajectories of all pseudo-edge pixels to filter out edge regions with stable morphological evolution characteristics. The specific method of consistency detection is to compare the continuity of the optical flow trajectory of each pseudo-edge point across all consecutive frames to determine whether there are trajectory interruptions, abrupt displacement changes, or directional jumps. For trajectories with abrupt changes or discontinuities, their evolutionary trajectories in phase information are further retrieved, and their score changes in the interferometric risk map are compared. If the trajectory is found to be discontinuous in the optical flow field, and the phase response exhibits asymmetric changes, with the interferometric score showing drastic fluctuations in the temporal dimension, then the edge point is marked as having temporal drift risk and is removed. By verifying the cross-frame optical flow consistency of all pseudo-edge points, pixels with unstable motion trajectories, inconsistent depths, or lacking multi-angle overlap support are gradually removed, thereby improving the temporal stability and spatial consistency of the final edge data. This process relies on the spatial viewpoint set and optical flow trajectory library established in the first two steps, achieving a precise pseudo-edge removal mechanism by fusing motion characteristics, depth structure, and phase response data.

[0032] After optical flow consistency detection and unstable edge point removal, all verified pseudo-edge regions were classified as real edge regions, and a real edge evidence dataset was constructed. This dataset includes the position coordinates, motion trajectory, depth estimate, optical flow consistency score, and phase stability evaluation index of each edge region across multiple image frames. It also includes the score evolution curve of each region in the interferometric risk map for further analysis of its robustness under different sampling conditions. To ensure the spatial continuity of the dataset, boundary stitching was performed on all verified edge regions, and their contour curves were fitted to the image plane with sub-pixel precision, generating corresponding edge morphology templates for use as edge constraint inputs in subsequent geometric reconstruction and correction parameter optimization stages. This dataset construction process not only improves the geometric realism and temporal consistency of the edge data but also provides a highly reliable edge structure reference for subsequent imaging structure reconstruction and interferometric effect elimination by introducing multi-view depth fusion and motion consistency evaluation mechanisms, effectively ensuring the overall image measurement accuracy and the stability of the reconstruction model.

[0033] The S400 reconstructs the imaging geometry based on a real edge evidence dataset, performs real-time adjustments to the camera's intrinsic and extrinsic parameters, and performs online inversion and recursive optimization of nonlinear distortion parameters to establish a self-consistent correction field covering the entire time series, thereby achieving alignment between the optical path system and the imaging coordinate system. After obtaining a dataset of real edge evidence with high temporal stability and spatial consistency, in order to accurately correct the imaging geometry of industrial close-range photogrammetric images, it is necessary to reconstruct the camera imaging model based on this edge dataset, dynamically adjust intrinsic and extrinsic parameters, optimize the effects of nonlinear distortion, and construct an integrated correction benchmark covering the entire time series to achieve precise alignment between the optical path propagation path and the imaging coordinate system. The specific steps are as follows: Based on the edge coordinates, optical flow trajectories, phase features, and depth values ​​contained in the real edge evidence dataset, a spatial backprojection operation is performed on the corresponding edge regions in each frame of the image. Combining the actual acquisition parameters and calibration benchmarks, the actual projective geometric relationship between the camera imaging center and spatial points is estimated. This step involves reconstructing the true spatial position of the edge points in the 3D calibration space and comparing it with the pixel projection coordinates on the image plane to obtain the projection offset of the imaging optical path in the current frame. After completing the backprojection of all edge points, the spatial deviation between them and the ideal projection model is statistically analyzed to construct an error distribution map of the current imaging frame. This error map describes the projection offset characteristics of each point in pixels, and the imaging geometry reconstruction task is performed using this map as input to establish the initial geometric structure model of the current image frame. This model is the core reference framework for subsequent correction processes.

[0034] Based on the initial geometric model, dynamic adjustments to the camera's intrinsic and extrinsic parameters are performed around the error distribution of all edge points in the image frame. Specifically, parameters reflecting the camera's imaging center position, lens focal length, optical axis offset, and principal point position are used as variables, and the camera's rotation matrix and translation vector are simultaneously optimized using external pose calibration data. During this optimization process, the spatial reprojection path provided in the real edge evidence dataset serves as a constraint, with the goal of minimizing the projection error of all edge points in the image frame. This allows for real-time adjustments to the camera's internal structure and external pose information, establishing a more accurate mapping relationship between the camera and the target object at the current sampling time. After this step, the intrinsic and extrinsic parameter set of the current frame image is used as the calibration output at the time node for subsequent processing stages. Simultaneously, the changing trends of each parameter over time are recorded to provide data support for subsequent recursion and interpolation.

[0035] After initial adjustments to camera parameters, a nonlinear distortion compensation mechanism is introduced to perform inverse modeling and recursive optimization of distortion phenomena caused by lens distortion, uneven lens curvature, or assembly errors. Specifically, for each frame's edge points, the radial offset and tangential displacement trends are extracted by comparing their theoretical projection trajectories with the pixel positions in the actual acquired image, thus constructing a nonlinear distortion offset model for the current frame. Subsequently, the distortion parameters of each frame are uniformly normalized in the time dimension, and a time-dependent distortion trend trajectory is constructed to establish the parameter evolution path. Combining the adjustments to camera intrinsic and extrinsic parameters from the previous stage, the distortion model undergoes recursive inversion processing to correct the nonlinear relationship between the optical path and the image plane, eliminating error accumulation caused by lens structure changes or temperature drift during imaging. During the recursive optimization process, the distortion compensation results for each frame are verified against the actual edge contour matching, and local reconstruction is performed in areas with low edge matching to improve overall geometric consistency.

[0036] After completing the above three processing steps, the corrected camera parameter set and nonlinear distortion model in each frame image are fused together to construct a self-consistent calibration field covering the entire sampling time series. This calibration field uses time frames as indices to record the complete parameter set applicable to each frame image at the imaging time, including the intrinsic parameter matrix, extrinsic parameter pose information, nonlinear distortion compensation model, and edge projection reconstruction accuracy score. To improve the accuracy of subsequent multi-frame image fusion and geometric model construction, this calibration field also introduces a spatial consistency evaluation mechanism. By comparing the spatial alignment errors of the same edge structure at different times, it dynamically adjusts the interpolation strategy and optimization weights between parameters, ensuring a high degree of consistency between the changes in the imaging optical path and the actual image features at different time nodes. During the image correction execution phase, all original image frames undergo refined geometric alignment processing based on the corresponding parameters provided by this calibration field, enabling each frame image to be accurately mapped to its true physical location in space. This ultimately achieves integrated alignment between the imaging optical path and the coordinate reference system, providing a unified and reliable geometric basis for subsequent high-precision measurement, 3D modeling, and error control.

[0037] The S500 performs dynamic control operations based on a self-consistent correction field. Through phase conjugate polarization rotation scanning combined with a reversible time grid rearrangement mechanism, it performs periodic reverse modulation on the energy distribution of interference pseudo-peaks, continuously suppresses the phase reversal region, and constructs a closed-loop control process for interference detection and interference source suppression. After constructing a self-consistent correction field covering the entire time sequence, to further enhance the adaptability of the imaging optical path to complex interference fields and actively suppress the spurious peak effect generated by highly reflective target surfaces, it is necessary to introduce a dynamic control mechanism based on phase manipulation and time reconstruction, establish a closed-loop control process for interference detection and interference source suppression, and achieve precise control and dynamic adjustment of the phase reversal region during the imaging stage. The specific steps are as follows: Based on the imaging parameters, optical path structure, and distortion correction model of each frame of image recorded in the self-consistent correction field at the imaging time, a spatial alignment reference required for the dynamic control framework is constructed. Specifically, according to the image frame index, all edge projection information, optical path distribution, camera intrinsic and extrinsic parameters, and nonlinear distortion compensation curves are jointly modeled in multiple dimensions to form an optical path control structure with temporal continuity and spatial consistency. This structure not only records the imaging geometric change trend of each pixel in different time periods but also marks the energy fluctuation pattern of the phase reversal region identified in the interferometric risk map throughout the entire time series. Based on this, known temporal characteristics of interference pseudo-peaks are retrieved, and positions prone to peak enhancement under periodic modulation are identified as target control nodes, providing stable triggering conditions for subsequent phase conjugate modulation.

[0038] After constructing the control reference structure, a phase reversal feature in the high-risk region is optically adjusted by introducing a phase conjugate polarization rotation scanning mechanism with dynamic response capabilities. In this step, a polarization rotation device with high-precision angle control is used to scan the incident beam in sections within the target region. By adjusting the polarization angle and polarization axis direction, the polarization state of the incident light is actively intervened, causing it to form a conjugate structure in the high-reflection region that is phase-complementary to the original reflection path. Simultaneously, a portion of the signal in the reflection path is guided into the interference suppression channel to shield local phase noise. During this process, the polarization rotation rate and direction are dynamically adjusted according to the optical path change trend provided by the preceding correction field, ensuring that the incident and reflected light form a mutually canceling state with anti-phase characteristics during actual imaging. This method uses the frequency change of the interference pseudo-peak as a reference standard, anchoring the control action at the critical point where phase superposition is most likely to occur, thereby achieving a source intervention effect on the interference energy.

[0039] Building upon optical-level polarization phase adjustment, a reversible temporal grid rearrangement mechanism is introduced to perform a temporal reverse reconstruction of the interference intensity evolution trajectory in the image frame sequence. This operation uses the energy change curves of spurious peak regions recorded in the interference risk map as input reference, reordering the original time frames according to the frequency of spurious peak occurrence and the phase jump period. Specifically, the time frame intervals in high interference frequency regions are weighted and shortened, allowing for a more densely controlled response in the time series, while maintaining a normal rhythm for stable frames, thereby compressing the accumulation period of interference peaks in the temporal dimension. During the reverse rearrangement process, inter-frame overlap interpolation and spatial compensation are performed for phase reversal regions, projecting the stable phase characteristics of known low-interference frames onto adjacent unstable frames. The temporal reconstruction accuracy is verified through optical flow consistency comparison and depth reconstruction checks. This mechanism, based on the theory of temporal reversibility, dynamically suppresses the interference propagation path by controlling the temporal grid structure of the image sequence, breaking the temporally continuous superposition chain upon which spurious peak formation depends.

[0040] By integrating the aforementioned phase conjugate polarization rotation adjustment and time grid reverse arrangement control mechanisms, a closed-loop control process for interference suppression is constructed. In this process, based on real-time parameters provided by the self-consistent correction field, before each image frame enters the acquisition stage, corresponding optical path control commands are triggered according to its predicted interference risk level. These commands include polarization angle preset, conjugate state configuration, and loading of time series fine-tuning rules. After the actual image acquisition is completed, based on the feedback grayscale distribution and phase response trend, the intensity of the pseudo-peaks in the interference-sensitive region of the current frame is automatically checked to ensure it meets the standard. The control parameters for the next frame are adjusted in conjunction with the response of the previous frame, forming a cyclical process of periodic response-feedback-re-control. During this process, the interference energy is gradually weakened to below the recognition threshold, and the original pseudo-peak structure is continuously suppressed in both spatial and temporal dimensions. This ultimately forms an imaging optical path adjustment mechanism with real-time adaptive capabilities, ensuring that even under high reflection and complex interference conditions, industrial close-range photogrammetric images with high geometric accuracy, stable texture boundaries, and continuous phase structures can still be output, providing a reliable image foundation for subsequent digital reconstruction and precision measurement.

[0041] This invention achieves dynamic perception and risk warning of image interference potential by constructing a temporally continuous anchor point sequence and interference risk map. Combined with phase polarization consistency auditing and pixel-by-pixel tracking, it accurately identifies and isolates pseudo-edge regions with phase reversal characteristics. Under a multi-view optical flow registration mechanism, it further eliminates unstable feature points disturbed by interference, thereby extracting highly reliable true edge evidence. Imaging geometry reconstruction and distortion optimization based on this evidence dataset enable precise full-time correction of imaging parameters, establishing an integrated alignment relationship between the optical path and the image. Furthermore, by employing phase conjugate polarization scanning and temporal grid rearrangement strategies, an active modulation and periodic suppression mechanism for interference energy is formed, constructing a closed-loop control process from interference detection to adaptive modulation, achieving full-process suppression and feedback adjustment of the interference source. The overall scheme possesses the synergistic advantages of dynamic perception, precise correction, and adaptive suppression, ultimately achieving accurate identification and stable elimination of interference pseudo-peaks in industrial close-up images, providing a solid data foundation and robust image assurance for high-precision 3D reconstruction and precision visual measurement.

[0042] This embodiment provides, for example Figure 2 The calibration system shown for industrial close-range photogrammetry images includes: The time-scaled optical path observation module is configured to construct a cross-time-scale optical path observation baseline under multi-time-scale sampling conditions, perform joint extraction operations of brightness distribution and phase gradient on each frame image, generate a time-continuous anchor point sequence, and construct an interference risk map based on the trend of optical path difference change, which is used as the dynamic constraint basis for feature recognition. The phase consistency audit module is configured to perform phase polarization consistency audit based on anchor point sequence and interferometric risk map. It performs pixel-by-pixel tracking of phase abrupt change regions in the brightness field, extracts interferometric pseudo-peaks with phase reversal characteristics, and outputs a pseudo-edge candidate dataset containing spatial location and phase state. The optical flow coupling registration module is configured to perform multi-view optical flow deep coupling registration based on the pseudo edge candidate dataset and phase audit results. It eliminates edge points with temporal drift through cross-frame optical flow consistency detection and transforms edge regions with stable morphological features into real edge evidence datasets. The geometric correction and reconstruction module is configured to reconstruct the imaging geometry based on the real edge evidence dataset, perform real-time adjustment of camera intrinsic and extrinsic parameters, perform online inversion and recursive optimization of nonlinear distortion parameters, establish a self-consistent correction field covering the entire time series, and achieve alignment between the optical path system and the imaging coordinate system. The interference suppression control module is configured to perform dynamic control operations based on the self-consistent correction field. Through phase conjugate polarization rotation scanning combined with a reversible time grid rearrangement mechanism, it performs periodic reverse modulation on the energy distribution of interference pseudo-peaks, continuously suppresses the phase reversal region, and constructs a closed-loop control process for interference detection and interference source suppression.

[0043] The correction method for industrial close-range photogrammetry images provided in this embodiment of the invention is implemented by the above-described correction system for industrial close-range photogrammetry images. For details of the specific methods and processes of the correction system for industrial close-range photogrammetry images, please refer to the embodiments of the correction method for industrial close-range photogrammetry images described above, which will not be repeated here.

[0044] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A correction method for industrial close-range photogrammetric images, characterized in that, Includes the following steps: S100 constructs a cross-timescale optical path observation baseline under multi-timescale sampling conditions, performs joint extraction operations of brightness distribution and phase gradient on each frame of image to generate anchor point sequence, and constructs an interferometric risk map based on the trend of optical path difference change, which is used as the dynamic constraint basis for feature recognition. S200 performs phase polarization consistency audit based on anchor point sequence and interferometric risk map, performs pixel-by-pixel tracking of phase abrupt change regions in brightness field, extracts interferometric pseudo-peaks with phase reversal characteristics, and outputs pseudo-edge candidate dataset; S300 performs multi-view optical flow deep coupling registration based on pseudo edge candidate dataset and phase audit results. It eliminates edge points with temporal drift through cross-frame optical flow consistency detection and transforms edge regions with stable morphological features into real edge evidence datasets. S400 reconstructs the imaging geometry based on a real edge evidence dataset, performs real-time adjustments to the camera's intrinsic and extrinsic parameters, and performs online inversion and recursive optimization of nonlinear distortion parameters to establish a self-consistent correction field covering the entire time series. The S500 performs dynamic control operations based on a self-consistent correction field. Through phase conjugate polarization rotation scanning combined with a reversible time grid rearrangement mechanism, it performs periodic reverse modulation on the energy distribution of interference pseudo-peaks, continuously suppressing the phase reversal region, and constructs a closed-loop control process for interference detection and interference source suppression.

2. The correction method for industrial close-range photogrammetry images according to claim 1, characterized in that, Step S100 includes: An imaging structure with spatial calibration reference is established under multi-timescale sampling conditions, and image frame sequences are acquired by configuring a multispectral imaging device with adjustable focal length. Based on the acquired image frame sequence, the joint feature data of brightness distribution and phase gradient are extracted, and the local gray-level continuity and phase change trend of pixels are analyzed to construct a joint feature point dataset. Based on the constructed joint feature point dataset, regions with consistent change characteristics in the time series are selected as anchor point candidate regions, and spatial back projection verification is performed in conjunction with the optical path observation baseline to determine the time anchor point sequence; Based on the time anchor sequence and optical path change trend, grayscale fluctuation and phase oscillation analysis are performed on non-anchor regions to construct an interference risk map for dynamic constraints and record the distribution of interference risk regions in the image.

3. The correction method for industrial close-range photogrammetry images according to claim 1, characterized in that, Step S200 includes: Based on the time anchor sequence and interference risk map, continuous pixel regions with high interference risk scores are selected as auditable regions within the pixel range of the image, and spatial proximity relationships with anchor points are established. The brightness change curve of each pixel in the audit area is compared with the brightness trajectory of the anchor point, and the grayscale response difference under different polarization directions is combined to identify areas with inconsistent phase. For each pixel in the phase inconsistency region, perform pixel-by-pixel time-series tracking, extract the brightness polarity reversal point and polarization response synchronization features, and determine whether there is a phase reversal phenomenon. Based on the judgment results, a pseudo-edge candidate dataset containing spatial location and phase state information is output, and boundary fitting and cross-validation with time anchor points are performed to form a pseudo-peak trajectory set with a clear structure.

4. The correction method for industrial close-range photogrammetry images according to claim 3, characterized in that, During pixel-by-pixel time series tracking, the phase reversal phenomenon is determined by the distribution frequency of brightness polarity reversal points, the spatial jump amplitude, and the synchronicity of polarization response changes. The corresponding pixel is marked as an interference pseudo-peak only when all three conditions are met.

5. The correction method for industrial close-range photogrammetry images according to claim 1, characterized in that, Step S300 includes: Based on the spatial location index and phase history information of the pseudo edge candidate dataset, the corresponding regions in the multi-view image sequence are retrieved, and the observation frame sequence with temporal relationship is constructed for subsequent analysis. For each observation frame sequence, optical flow field is tracked frame by frame in the pseudo-edge region to establish pixel motion trajectories between consecutive image frames, and disparity matching is introduced to calculate the spatial depth consistency of each pixel. Based on the optical flow tracing results, cross-frame consistency detection is performed on all pseudo-edge points. Combining the phase response trajectory and the change in interference score, unstable edge points with temporal drift and spatial jump are eliminated. All edge regions that pass the consistency verification are constructed into a real edge evidence dataset, and spatial location, depth estimation and contour morphology information are extracted to provide highly reliable input for subsequent geometric reconstruction.

6. The correction method for industrial close-range photogrammetry images according to claim 5, characterized in that, In the process of constructing a real edge evidence dataset, subpixel-level boundary fitting is performed on the edge regions that pass the consistency verification, and edge morphology templates are generated to serve as edge constraint inputs in subsequent imaging geometry reconstruction.

7. The correction method for industrial close-range photogrammetry images according to claim 1, characterized in that, Step S400 includes: Based on the edge coordinates, optical flow trajectory, phase features and depth values ​​in the real edge evidence dataset, a spatial back projection operation is performed on each frame of the image to establish the projection geometry of the imaging optical path; An error distribution map is constructed based on the projection geometry, and dynamic adjustment operations are performed on the camera's intrinsic and extrinsic parameters to optimize the camera's imaging center position, focal length, optical axis offset, and external attitude information. After adjusting the camera parameters, nonlinear distortion parameters are modeled and recursively optimized based on the theoretical trajectory and actual position of the edge points to compensate for radial and tangential distortion errors. The camera parameters of each frame are fused with the distortion model to construct a self-consistent correction field covering the entire time series, which is used to guide the geometric alignment and optical path consistency correction of all image frames in space.

8. The correction method for industrial close-range photogrammetry images according to claim 7, characterized in that, In the recursive optimization process of nonlinear distortion parameters, the distortion model is locally reconstructed based on the matching error of edge contours in each frame image, and the parameter weights are dynamically adjusted using a spatial consistency evaluation mechanism to improve the geometric correction accuracy.

9. The correction method for industrial close-range photogrammetry images according to claim 1, characterized in that, Step S500 includes: Based on the imaging parameters, optical path structure and distortion correction model recorded in the self-consistent correction field, an optical path control structure with temporal continuity and spatial consistency is constructed, and a target control node with phase reversal characteristics is identified. By using a phase conjugate polarization rotation scanning mechanism, the polarization state of the incident light is adjusted within the high-reflection region to form a structure that is phase-complementary to the reflection path, thereby interfering with the phase superposition process of the interference pseudo-peaks. By reversible temporal raster rearrangement mechanism, the evolution trajectory of interference intensity in the image frame sequence is reconstructed, the inter-frame temporal structure is adjusted to weaken the accumulation of pseudo-peak energy, and a temporal dimension interference control strategy is constructed. Based on the aforementioned dual phase and time adjustment method, a closed-loop control process for interference detection and interference source suppression is constructed. The control parameters are set before image acquisition and the control command for the next frame is updated based on the response feedback after acquisition.

10. A correction system for industrial close-range photogrammetric images, used to implement the correction method for industrial close-range photogrammetric images according to any one of claims 1-9, characterized in that, include: The time-scaled optical path observation module is configured to construct a cross-time-scale optical path observation baseline under multi-time-scale sampling conditions, perform joint extraction operations of brightness distribution and phase gradient on each frame image, generate an anchor point sequence, and construct an interference risk map based on the trend of optical path difference change, which is used as the dynamic constraint basis for feature recognition. The phase consistency audit module is configured to perform phase polarization consistency audit based on anchor point sequence and interferometric risk map, track phase change regions in brightness field pixel by pixel, extract interferometric pseudo-peaks with phase reversal characteristics, and output pseudo-edge candidate dataset; The optical flow coupling registration module is configured to perform multi-view optical flow deep coupling registration based on the pseudo edge candidate dataset and phase audit results. It eliminates edge points with temporal drift through cross-frame optical flow consistency detection and transforms edge regions with stable morphological features into real edge evidence datasets. The geometric correction and reconstruction module is configured to reconstruct the imaging geometry based on the real edge evidence dataset, perform real-time adjustment of camera intrinsic and extrinsic parameters, and perform online inversion and recursive optimization of nonlinear distortion parameters to establish a self-consistent correction field covering the entire time series. The interference suppression control module is configured to perform dynamic control operations based on the self-consistent correction field. Through phase conjugate polarization rotation scanning combined with a reversible time grid rearrangement mechanism, it performs periodic reverse modulation on the energy distribution of interference pseudo-peaks, continuously suppresses the phase reversal region, and constructs a closed-loop control process for interference detection and interference source suppression.

Images