Industrial endoscope measurement system based on binocular image acquisition and ai intelligent correction

Abstract

This invention discloses an industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction, belonging to the field of endoscope measurement technology. The invention plans a binocular acquisition path on the measurement target, retrieves the measurement images of the target acquired by the industrial endoscope along the binocular acquisition path, maps the measurement images acquired by each industrial endoscope in a spatiotemporal order onto a three-dimensional mapping space, sequentially stitches the image feature extraction results of each acquisition path in a spatiotemporal order to obtain the target feature path, performs multi-angle overlap of each target feature path in a spatiotemporal order, obtains multi-angle consensus features based on the multi-angle overlap results, retrieves each multi-angle consensus feature to traverse the target feature path, determines whether there is a deviation in the target feature path of each industrial endoscope based on the traversal results, and adjusts the industrial endoscopes with deviations based on the determination results.

Description

Technical Field

[0001] This invention relates to the field of endoscopic measurement technology, and more specifically to an industrial endoscopic measurement system based on binocular image acquisition and AI intelligent correction. Background Technology

[0002] In the field of industrial inspection, especially for defect measurement and morphology analysis inside complex cavities such as aero-engines and precision pipelines, industrial endoscopes are an indispensable core tool. Traditional industrial endoscopic measurements mainly rely on monocular imaging and manual interpretation. The measurement accuracy is severely limited by the experience level of the inspectors, and single-view images cannot provide accurate depth information, making it impossible to achieve three-dimensional quantification of complex curved surfaces or hidden defects. Some existing binocular or three-dimensional endoscope systems acquire parallax information through physical dual lenses for measurement. However, within the narrow, unstructured industrial cavities, how to plan an effective binocular collaborative acquisition path, how to robustly extract and match features from multi-view, multi-time-series images, and how to verify the equipment's own status online to avoid systematic errors introduced by lens contamination, mechanical deformation, etc., remain pressing technical challenges. These problems result in limitations in the practical application of existing systems, such as cumbersome measurement processes, low automation, and unstable result reliability.

[0003] While computer vision and artificial intelligence technologies have provided new methods for image feature extraction and matching, directly applying general algorithms often yields poor results in the specific scenario of industrial endoscopy. The uneven lighting, weak texture, and complex background inside industrial cavities make general feature extraction algorithms prone to failure. Furthermore, the continuous movement of the endoscope generates image sequences containing spatiotemporal information, rather than simple pairs of images. Traditional 3D reconstruction methods struggle to effectively utilize this sequence information for high-precision path-level 3D reconstruction and self-calibration. Therefore, an industrial endoscopy measurement system based on binocular image acquisition and AI intelligent correction is needed. Summary of the Invention

[0004] The purpose of this invention is to provide an industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction, so as to solve the problems of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: An industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction includes a binocular acquisition planning module, an AI feature extraction module, and a verification and comparison module. The binocular acquisition planning module is used to retrieve the initial measurement image of the measurement target acquired by the industrial endoscope, and send the initial measurement image to the AI ​​feature extraction module. Then, based on the measurement recognition texture on the initial measurement image, the binocular acquisition path is planned on the measurement target, and the result measurement image of the measurement target acquired by the industrial endoscope along the binocular acquisition path is retrieved. The AI ​​feature extraction module is used to mark the measurement recognition texture on the initial measurement image and establish a three-dimensional mapping space according to the binocular acquisition path. The measurement images acquired by each industrial endoscope are mapped into the three-dimensional mapping space in a spatiotemporal order. Then, image features are extracted from the measurement images from the same industrial endoscope. The image feature extraction results of each acquisition path are sequentially spliced ​​in a spatiotemporal order to obtain the target feature path. The verification and comparison module is used to overlap each target feature path in a spatiotemporal order from multiple angles, obtain multi-angle consensus features based on the multi-angle overlap results, retrieve each multi-angle consensus feature pair to traverse the target feature path, determine whether there is a deviation in the target feature path of each industrial endoscope based on the traversal results, and adjust the industrial endoscope with deviation based on the judgment results.

[0006] Furthermore, the initial measurement image acquisition process includes: Several industrial endoscopes of the same specification are deployed at the industrial endoscope measurement site. The industrial endoscopes are equipped with a high-definition imaging module, an angle adjustment gimbal, a height adjustment slide, and a motion trajectory sensor. After fixing the target to be measured at the inspection station, the high-definition imaging module of the industrial endoscope is activated to control a single industrial endoscope to perform a full-area scan of the target and acquire the initial measurement image of the target. The initial measurement image covers all areas of the target to be measured.

[0007] Furthermore, the process of marking the measurement recognition texture on the initial measurement image includes: The measurement and identification texture is a unique and distinctive structural texture on the target surface. Based on the pixel characteristics of the measurement and identification texture, the measurement and identification texture is accurately marked on the initial measurement image. During the marking process, the edge pixels of the measurement and identification texture are used as a reference, and a marking width of 5 or more pixels is set to mark the measurement and identification texture on the initial measurement image.

[0008] Furthermore, the binocular acquisition path planning process includes: Using texture measurement and recognition as anchor points, the area to be measured of the target is divided into several acquisition sub-regions. The number of acquisition sub-regions is equal to the number of industrial endoscopes, ensuring that each device is responsible for an independent acquisition sub-region. Then, an acquisition path is planned for each acquisition sub-region. The trajectory of the acquisition path is a continuous curve or polyline, covering all the positions to be measured in the corresponding acquisition sub-region. Finally, collision detection is performed on all acquisition paths. Through three-dimensional spatial simulation technology, the process of multiple industrial endoscopes moving synchronously along each acquisition path is simulated. If the movement trajectories are detected to intersect or overlap, the spatial parameters of the corresponding acquisition path are adjusted according to the intersection or overlap position until the movement trajectories of all industrial endoscopes do not collide and the acquisition ranges do not overlap, thus forming a binocular acquisition path composed of multiple acquisition paths. After the binocular acquisition path is planned, multiple industrial endoscopes are used to acquire the measurement images of the target along the preset binocular acquisition path.

[0009] Furthermore, the process of mapping the measurement images acquired by various industrial endoscopes into a three-dimensional mapping space in spatiotemporal order includes: A three-dimensional mapping space is constructed based on the three-dimensional spatial parameters of the binocular acquisition path issued by the binocular acquisition planning module. The coordinate system of the three-dimensional mapping space is consistent with the actual spatial coordinate system of the measurement target, and the spatial ratio is 1:1 to ensure the accuracy of image mapping. After each industrial endoscope completes the acquisition of the result measurement images, it sends the continuous frames of result measurement images to the AI ​​feature extraction module in real time. The AI ​​feature extraction module maps the result measurement images of each industrial endoscope to the corresponding positions in the three-dimensional mapping space according to the spatiotemporal synchronization principle. That is, according to the acquisition time, acquisition position and acquisition angle of the industrial endoscope, each frame of result measurement image is mapped to the coordinate point in the three-dimensional mapping space that matches the actual acquisition position, so as to ensure that the spatial position of the image corresponds one-to-one with the actual acquisition position. For continuous frame measurement images from the same industrial endoscope, the AI ​​feature extraction module performs stitching and overlapping processing according to the acquisition time sequence. During the stitching process, the overlapping area of ​​two adjacent frames is used as a reference, and the pixel matching algorithm is used to accurately match the pixels in the overlapping area to achieve seamless stitching and form a continuous image sequence corresponding to a single industrial endoscope.

[0010] Furthermore, the process of sequentially concatenating the image feature extraction results from each acquisition path in spatiotemporal order to obtain the target feature path includes: Feature pointers are set in a three-dimensional mapping space. The feature pointers are feature detection units with pixel-level recognition capabilities. They contain feature image segments of the measurement target obtained through the Internet. The feature image segments contained in each feature pointer are not completely the same. That is, the feature image segments of any two feature pointers are either the same or completely different. Their motion trajectory is consistent with the acquisition path of the industrial endoscope, and their motion speed is synchronized with the image acquisition frame rate. The feature pointer extracts image features point by point and region by region from the continuous image sequence after splicing and overlapping, according to the acquisition time sequence of the measured images. For a single acquisition path, the AI ​​feature extraction module linearly splices all the feature points extracted by the feature pointer according to the spatiotemporal order of image acquisition to form a continuous target feature path with spatial coordinate information.

[0011] Furthermore, the process of overlapping the various target feature paths in a spatiotemporal order from multiple angles and obtaining multi-angle consensus features based on the results of the multi-angle overlap includes: After receiving multiple target feature paths sent by the AI ​​feature extraction module, the verification and comparison module performs multi-dimensional and multi-angle overlapping matching of all target feature paths in the three-dimensional mapping space according to the spatiotemporal order of image acquisition. During the overlapping matching process, the feature points of different target feature paths at the same acquisition time point are matched with time as the axis, and the spatial coordinates and feature parameters of each feature point are compared to see if they are consistent. At the same time, the overall trajectories of different target feature paths are compared to analyze the overlap and deviation of the trajectories. By using multi-dimensional and multi-angle overlapping matching, the verification and comparison module extracts the common features between different target feature paths, which are denoted as multi-angle consensus features. The multi-angle consensus features are the same feature positions that exist in the same spatiotemporal order of the target feature paths corresponding to different industrial endoscopes. The spatial coordinates, texture features, and morphological parameters of the feature positions are consistent in all target feature paths.

[0012] Furthermore, the process of retrieving various multi-angle consensus feature pairs to traverse the target feature path, and determining whether there are deviations in the target feature paths of each industrial endoscope based on the traversal results, includes: All multi-angle consensus features are integrated to form a consensus feature set. Based on the consensus feature set, all target feature paths are traversed and verified. During the traversal and verification process, the feature verification pointer traverses the trajectory of the target feature path point by point. The spatial coordinates and feature parameters of each feature point are compared with the multi-angle consensus features of the corresponding spatiotemporal location in the consensus feature set. The deviation value between the two is calculated, and a deviation threshold is set. If, after traversal and verification, the deviation values ​​of all feature points on the target feature path are less than the deviation threshold, then it is determined that the industrial endoscope acquisition corresponding to the target feature path has no deviation. If the deviation value of a feature point on the target feature path is greater than or equal to the deviation threshold, it is determined that there is a sampling deviation on the target feature path, and the corresponding industrial endoscope sampling accuracy does not meet the requirements. The verification and comparison module analyzes the causes of the acquisition deviation based on the deviation value, deviation direction and spatial distribution of the deviation feature points, and determines the corresponding debugging parameters. The debugging content mainly includes the acquisition angle and acquisition height of the industrial endoscope. If the spatial coordinates of the deviation feature point are offset in the horizontal direction, it is determined to be a sampling angle deviation. The verification and comparison module sends a debugging command to the angle adjustment pan-tilt unit of the industrial endoscope. Each time the debugging command is executed, the result measurement image of the corresponding area is re-acquired, and a new target feature path is generated.

[0013] Furthermore, if the number of consecutive executions of the debugging command reaches a preset threshold, and the deviation feature points on the target feature path still have deviations exceeding the deviation threshold, and the deviation value does not show an obvious decreasing trend, then it is determined that the acquisition deviation is not caused by improper acquisition parameter settings, but by the hardware defects of the industrial endoscope itself. Then, the verification and comparison module uses the spatial coordinates, deviation direction and deviation value of the deviation feature points, combined with the current acquisition angle and height of the corresponding industrial endoscope, to obtain the source location of the deviation through three-dimensional spatial inverse operation, that is, the defect location of the industrial endoscope.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention uses the measurement and recognition of textures based on the initial image to intelligently plan the optimal acquisition path, guide the endoscope to perform systematic scanning, and robustly extract and map the features of the time-series acquired image sequences. The images are then stitched together to form a continuous and complete target feature path, thereby constructing a high-precision three-dimensional morphology of the measured target. Compared with the limitations of traditional monocular measurement or fixed dual-target positioning, this invention achieves automated three-dimensional data acquisition in unstructured environments, improves the accuracy and efficiency of defect size measurement and morphology analysis, and reduces reliance on skilled operators.

[0015] 2. This invention compares the target feature paths obtained by different endoscopes or different paths from multiple angles, extracts the consensus features from multiple angles as a benchmark, and then traverses and verifies the consistency of each feature path. This enables automatic identification and location of measurement deviations caused by abnormal conditions of a single endoscope lens, mechanical errors, etc., and triggers debugging prompts. Attached Figure Description

[0016] 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.

[0017] Figure 1This is a system block diagram of the industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction as described in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 As shown, the industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction includes a binocular acquisition planning module, an AI feature extraction module, and a verification and comparison module. The binocular acquisition planning module is used to retrieve the initial measurement image of the measurement target acquired by the industrial endoscope, and send the initial measurement image to the AI ​​feature extraction module. Then, based on the measurement recognition texture on the initial measurement image, the binocular acquisition path is planned on the measurement target, and the result measurement image of the measurement target acquired by the industrial endoscope along the binocular acquisition path is retrieved. The AI ​​feature extraction module is used to mark the measurement recognition texture on the initial measurement image and establish a three-dimensional mapping space according to the binocular acquisition path. The measurement images acquired by each industrial endoscope are mapped into the three-dimensional mapping space in a spatiotemporal order. Then, image features are extracted from the measurement images from the same industrial endoscope. The image feature extraction results of each acquisition path are sequentially spliced ​​in a spatiotemporal order to obtain the target feature path. The verification and comparison module is used to overlap each target feature path in a spatiotemporal order from multiple angles, obtain multi-angle consensus features based on the multi-angle overlap results, retrieve each multi-angle consensus feature pair to traverse the target feature path, determine whether there is a deviation in the target feature path of each industrial endoscope based on the traversal results, and adjust the industrial endoscope with deviation based on the judgment results.

[0020] Furthermore, the working principle of the present invention will be illustrated below through embodiments: Several industrial endoscopes of the same specifications are deployed at the industrial endoscope measurement site. The industrial endoscopes are equipped with a high-definition imaging module, an angle adjustment gimbal, a height adjustment slide, and a motion trajectory sensor. The resolution of the high-definition imaging module is no less than 2560×1440, and the frame rate is set to 30Hz. The adjustment range of the angle adjustment gimbal is 0-180°, and the adjustment stroke of the height adjustment slide is 0-50cm, which can realize fine adjustment of the acquisition angle and height. The motion trajectory sensor is connected to the binocular acquisition planning module to provide real-time feedback on the motion trajectory of the industrial endoscope, ensuring that it moves strictly along the preset binocular acquisition path.

[0021] At the same time, the industrial component to be inspected is used as the measurement target and fixed at the inspection station to ensure that the measurement target does not shift or shake during the data acquisition process, thus ensuring the accuracy of the data acquisition. The high-definition imaging module of the industrial endoscope is connected to the image receiver of the binocular acquisition planning module, and the angle adjustment gimbal, height adjustment slide, and motion trajectory sensor are all connected to the control terminal of the binocular acquisition planning module. After fixing the target to be measured at the inspection station, the high-definition imaging module of the industrial endoscope is activated to control a single industrial endoscope to perform a full-area scan of the target and acquire the initial measurement image of the target. The initial measurement image covers all areas of the target to be measured, ensuring that the overall structure and surface texture of the target can be clearly identified. The binocular acquisition planning module sends the initial measurement image to the AI ​​feature extraction module in real time. The AI ​​feature extraction module then identifies and locates the measurement recognition texture on the initial measurement image. The measurement recognition texture is a unique and distinctive structural texture on the surface of the measurement target, such as weld seams, grooves, bolt holes, and stepped surfaces. The process of identifying and locating the measurement recognition texture on the initial measurement image includes: accurately marking the measurement recognition texture on the initial measurement image based on the pixel features of the measurement recognition texture. During the marking process, the edge pixels of the measurement recognition texture are used as a reference, and a marking width of 5 or more pixels is set to mark the measurement recognition texture on the initial measurement image. After the measurement recognition texture of all initial measurement images is marked, the AI ​​feature extraction module sends the initial measurement image to the binocular acquisition planning module. The binocular acquisition planning module plans the binocular acquisition path on the area to be measured of the target based on the spatial distribution and morphological features of the measurement recognition texture on the initial measurement image and the overall structure of the measurement target, combined with the number of industrial endoscopes deployed. The binocular acquisition path planning process includes: first, using the measured and identified texture as the anchor point, dividing the area to be measured of the target into several acquisition sub-regions. The number of acquisition sub-regions is equal to the number of industrial endoscopes, ensuring that each device is responsible for an independent acquisition sub-region; then, planning an acquisition path for each acquisition sub-region. The trajectory of the acquisition path is a continuous curve or polyline, covering all the positions to be measured in the corresponding acquisition sub-region. Finally, collision detection is performed on all acquisition paths. Through three-dimensional spatial simulation technology, the process of multiple industrial endoscopes moving synchronously along each acquisition path is simulated. If the movement trajectories are detected to intersect or overlap, the spatial parameters of the corresponding acquisition path are adjusted according to the intersection or overlap position until the movement trajectories of all industrial endoscopes do not collide and the acquisition ranges do not overlap, thus forming a binocular acquisition path composed of multiple acquisition paths. After the binocular acquisition path planning is completed, the binocular acquisition planning module sends the three-dimensional spatial parameters of the binocular acquisition path, including the coordinates of the start point, end point, and inflection points of the acquisition path, as well as the movement direction and speed of each segment of the acquisition path, to the corresponding industrial endoscopes. At the same time, it sends a synchronous acquisition command to all industrial endoscopes, calling up multiple industrial endoscopes to synchronously acquire images of the corresponding acquisition sub-regions of the measurement target along the preset binocular acquisition path, obtaining continuous frame measurement images. The acquisition angle of each industrial endoscope is determined by the spatial parameters of the acquisition path, always remaining different and covering the entire measurement area.

[0022] Furthermore, the AI ​​feature extraction module constructs a three-dimensional mapping space based on the three-dimensional spatial parameters of the binocular acquisition path issued by the binocular acquisition planning module. The coordinate system of the three-dimensional mapping space is consistent with the actual spatial coordinate system of the measurement target, and the spatial ratio is 1:1 to ensure the accuracy of image mapping. After each industrial endoscope completes the acquisition of the result measurement images, it sends the continuous frames of result measurement images to the AI ​​feature extraction module in real time. The AI ​​feature extraction module maps the result measurement images of each industrial endoscope to the corresponding positions in the three-dimensional mapping space according to the spatiotemporal synchronization principle. That is, according to the acquisition time, acquisition position and acquisition angle of the industrial endoscope, each frame of result measurement image is mapped to the coordinate point in the three-dimensional mapping space that matches the actual acquisition position, so as to ensure that the spatial position of the image corresponds one-to-one with the actual acquisition position. For continuous frame measurement images from the same industrial endoscope, the AI ​​feature extraction module performs splicing and overlapping processing according to the acquisition time sequence. During the splicing process, the overlapping area of ​​two adjacent frames is used as a reference, and the pixel matching algorithm is used to accurately match the pixels in the overlapping area to achieve seamless splicing and form a continuous image sequence corresponding to a single industrial endoscope. After the continuous image sequence is stitched together, the AI ​​feature extraction module sets feature pointers in the three-dimensional mapping space. The feature pointers are feature detection units with pixel-level recognition capabilities. They contain feature image segments of the measurement target obtained through the Internet. The feature image segments contained in each feature pointer are not completely the same. That is, the feature image segments of any two feature pointers are either the same or completely different. Their motion trajectory is consistent with the acquisition path of the industrial endoscope, and their motion speed is synchronized with the image acquisition frame rate. The feature pointer extracts image features point by point and region by region from the continuous image sequence after splicing and overlapping, according to the acquisition time sequence of the measured images. The extracted features include texture features, edge features, contour features, geometric size features, etc. of the measured target surface. At the same time, it records the spatial coordinates, pixel gray value, texture direction and other parameters of each feature point in the three-dimensional mapping space. For a single acquisition path, the AI ​​feature extraction module linearly stitches together all the feature points extracted by the feature pointer according to the spatiotemporal order of image acquisition to form a continuous target feature path with spatial coordinate information. Multiple industrial endoscopes generate multiple target feature paths. After integrating all the target feature paths, the AI ​​feature extraction module sends them to the verification and comparison module in real time.

[0023] Furthermore, after receiving multiple target feature paths sent by the AI ​​feature extraction module, the verification and comparison module performs multi-dimensional and multi-angle overlapping matching of all target feature paths in the three-dimensional mapping space according to the spatiotemporal order of image acquisition. During the overlapping matching process, with time as the axis, the feature points of different target feature paths at the same acquisition time point are matched, and the spatial coordinates and feature parameters of each feature point are compared to see if they are consistent. At the same time, the overall trajectories of different target feature paths are compared to analyze the overlap and deviation of the trajectories. Through multi-dimensional and multi-angle overlapping matching, the verification and comparison module extracts common features between different target feature paths, which are denoted as multi-angle consensus features. The multi-angle consensus features are the same feature positions that exist in the same spatiotemporal order of the target feature paths corresponding to different industrial endoscopes. The spatial coordinates, texture features, and morphological parameters of the feature positions are consistent in all target feature paths. They are determined by the actual structural features of the measured target and are not affected by the acquisition parameters of the industrial endoscope. Therefore, they become the core benchmark for judging the acquisition accuracy of each industrial endoscope. The number of multi-angle consensus features is positively correlated with the number of iconic measurement and identification textures on the surface of the measured target. The more iconic textures there are, the richer the multi-angle consensus features are, and the higher the judgment accuracy is. The verification and comparison module integrates all extracted multi-angle consensus features to form a consensus feature set. Based on the consensus feature set, it performs full path traversal verification on all target feature paths. During the traversal verification process, the feature verification pointer traverses the trajectory of the target feature path point by point, comparing the spatial coordinates and feature parameters of each feature point with the multi-angle consensus features of the corresponding spatiotemporal location in the consensus feature set, calculating the deviation value between the two, and setting a deviation threshold. The deviation threshold is set according to the accuracy requirements of industrial measurement, generally 0.01-0.05mm. If, after traversal and verification, the deviation values ​​of all feature points on the target feature path are less than the deviation threshold, then it is determined that the industrial endoscope acquisition corresponding to the target feature path has no deviation and the acquisition accuracy meets the requirements. If the deviation value of a feature point on the target feature path is greater than or equal to the deviation threshold, it is determined that there is a sampling deviation on the target feature path, and the corresponding industrial endoscope sampling accuracy does not meet the requirements. The verification and comparison module analyzes the causes of the acquisition deviation based on the deviation value, deviation direction and spatial distribution of the deviation feature points, and determines the corresponding debugging parameters. The debugging content mainly includes the acquisition angle and acquisition height of the industrial endoscope. If the spatial coordinates of the deviation feature point are offset in the horizontal direction, it is determined to be an acquisition angle deviation. The verification and comparison module sends a debugging command to the angle adjustment pan-tilt unit of the industrial endoscope to fine-tune the rotation angle of the pan-tilt unit in 0.1° increments. Each time the debugging command is executed, the industrial endoscope is controlled to reacquire the measurement image of the corresponding area, generate a new target feature path, and perform traversal verification again until the deviation value is less than the deviation threshold. If the spatial coordinates of the deviation feature point are offset in the vertical direction, it is determined to be an acquisition height deviation. The verification and comparison module sends a debugging command to the height adjustment slide of the industrial endoscope to fine-tune the height of the slide in 0.05cm increments. Similarly, after each adjustment, the acquisition and verification are re-acquired until the deviation value is less than the deviation threshold. If the number of consecutive executions of the debugging command reaches the preset threshold (usually 10 times), and the deviation feature points on the target feature path still have deviations exceeding the deviation threshold, and the deviation value does not show an obvious decreasing trend, then it is determined that the acquisition deviation is not caused by improper acquisition parameter settings, but by the hardware defects of the industrial endoscope itself. Then, the verification and comparison module calculates the source location of the deviation, i.e. the defect location of the industrial endoscope, based on the spatial coordinates, deviation direction and deviation value of the deviation feature points, combined with the current acquisition angle, height and other parameters of the corresponding industrial endoscope, through three-dimensional spatial inverse operation. The defects mainly include lens distortion of the high-definition imaging module, transmission clearance of the angle adjustment gimbal, positioning error of the height adjustment slide, and detection deviation of the motion trajectory sensor. The verification and comparison module visualizes and records the specific information of the defect locations (such as the lens, gimbal, etc.), providing accurate positioning basis for the subsequent maintenance, calibration and repair of industrial endoscopes, ensuring that the equipment can restore the acquisition accuracy in time and guarantee the accuracy of industrial measurements.

[0024] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction, characterized in that, It includes a binocular acquisition planning module, an AI feature extraction module, and a verification and comparison module; The binocular acquisition planning module is used to retrieve the initial measurement image of the measurement target acquired by the industrial endoscope, and send the initial measurement image to the AI ​​feature extraction module. Then, based on the measurement recognition texture on the initial measurement image, the binocular acquisition path is planned on the measurement target, and the result measurement image of the measurement target acquired by the industrial endoscope along the binocular acquisition path is retrieved. The AI ​​feature extraction module is used to mark the measurement recognition texture on the initial measurement image and establish a three-dimensional mapping space according to the binocular acquisition path. The measurement images acquired by each industrial endoscope are mapped into the three-dimensional mapping space in a spatiotemporal order. Then, image features are extracted from the measurement images from the same industrial endoscope. The image feature extraction results of each acquisition path are sequentially spliced ​​in a spatiotemporal order to obtain the target feature path. The verification and comparison module is used to overlap each target feature path in a spatiotemporal order from multiple angles, obtain multi-angle consensus features based on the multi-angle overlap results, retrieve each multi-angle consensus feature pair to traverse the target feature path, determine whether there is a deviation in the target feature path of each industrial endoscope based on the traversal results, and adjust the industrial endoscope with deviation based on the judgment results.

2. The industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction according to claim 1, characterized in that, The acquisition process of the initial measurement image includes: Several industrial endoscopes of the same specification are deployed at the industrial endoscope measurement site. The industrial endoscopes are equipped with a high-definition imaging module, an angle adjustment gimbal, a height adjustment slide, and a motion trajectory sensor. After fixing the target to be measured at the inspection station, the high-definition imaging module of the industrial endoscope is activated to control a single industrial endoscope to perform a full-area scan of the target and acquire the initial measurement image of the target. The initial measurement image covers all areas of the target to be measured.

3. The industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction according to claim 2, characterized in that, The process of marking measurement recognition textures on the initial measurement image includes: The measurement and identification texture is a unique and distinctive structural texture on the target surface. Based on the pixel characteristics of the measurement and identification texture, the measurement and identification texture is accurately marked on the initial measurement image. During the marking process, the edge pixels of the measurement and identification texture are used as a reference, and a marking width of 5 or more pixels is set to mark the measurement and identification texture on the initial measurement image.

4. The industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction according to claim 3, characterized in that, The binocular acquisition path planning process includes: Using texture measurement and recognition as the anchor point, the area to be measured of the target is divided into several acquisition sub-regions. The number of acquisition sub-regions is equal to the number of industrial endoscopes. Then, an acquisition path is planned for each acquisition sub-region. The trajectory of the acquisition path is a continuous curve or polyline, covering all the positions to be measured in the corresponding acquisition sub-region. Finally, collision detection is performed on all acquisition paths. Through three-dimensional spatial simulation technology, the process of multiple industrial endoscopes moving synchronously along each acquisition path is simulated. If the movement trajectories are detected to intersect or overlap, the spatial parameters of the corresponding acquisition path are adjusted according to the intersection or overlap position until the movement trajectories of all industrial endoscopes do not collide and the acquisition ranges do not overlap, thus forming a binocular acquisition path composed of multiple acquisition paths. After the binocular acquisition path is planned, multiple industrial endoscopes are used to acquire the measurement images of the target along the preset binocular acquisition path.

5. The industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction according to claim 4, characterized in that, The process of mapping the measurement images acquired by various industrial endoscopes into a three-dimensional mapping space in spatiotemporal order includes: A three-dimensional mapping space is constructed based on the three-dimensional spatial parameters of the binocular acquisition path issued by the binocular acquisition planning module. The coordinate system of the three-dimensional mapping space is consistent with the actual spatial coordinate system of the measurement target. After each industrial endoscope completes the acquisition of the result measurement images, it sends the continuous frames of result measurement images to the AI ​​feature extraction module in real time. The AI ​​feature extraction module maps the result measurement images of each industrial endoscope to the corresponding positions in the three-dimensional mapping space according to the principle of spatiotemporal synchronization. That is, according to the acquisition time, acquisition position and acquisition angle of the industrial endoscope, each frame of result measurement image is mapped to the coordinate point in the three-dimensional mapping space that matches the actual acquisition position. For continuous frame measurement images from the same industrial endoscope, the AI ​​feature extraction module performs splicing and overlapping processing according to the acquisition time sequence. During the splicing process, the overlapping area of ​​two adjacent frames is used as a reference to accurately match the pixels in the overlapping area, achieving seamless splicing and forming a continuous image sequence corresponding to a single industrial endoscope.

6. The industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction according to claim 5, characterized in that, The process of sequentially concatenating the image feature extraction results from each acquisition path in spatiotemporal order to obtain the target feature path includes: Feature pointers are set in a three-dimensional mapping space. The feature pointers are feature detection units with pixel-level recognition capabilities. They contain feature image segments of the measurement target obtained through the Internet. The feature image segments contained in each feature pointer are not completely the same. That is, the feature image segments of any two feature pointers are either the same or completely different. Their motion trajectory is consistent with the acquisition path of the industrial endoscope, and their motion speed is synchronized with the image acquisition frame rate. The feature pointer extracts image features point by point and region by region from the continuous image sequence after splicing and overlapping, according to the acquisition time sequence of the measured images. For a single acquisition path, the AI ​​feature extraction module linearly splices all the feature points extracted by the feature pointer according to the spatiotemporal order of image acquisition to obtain the target feature path.

7. The industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction according to claim 6, characterized in that, The process of overlapping various target feature paths in spatiotemporal order from multiple angles and obtaining multi-angle consensus features based on the results of the multi-angle overlap includes: After receiving multiple target feature paths sent by the AI ​​feature extraction module, the verification and comparison module performs multi-dimensional and multi-angle overlapping matching of all target feature paths in the three-dimensional mapping space according to the spatiotemporal order of image acquisition. During the overlapping matching process, the feature points of different target feature paths at the same acquisition time point are matched with time as the axis, and the spatial coordinates and feature parameters of each feature point are compared to see if they are consistent. At the same time, the overall trajectories of different target feature paths are compared to analyze the overlap and deviation of the trajectories. By using multi-dimensional and multi-angle overlapping matching, the verification and comparison module extracts the common features between different target feature paths, which are denoted as multi-angle consensus features. The multi-angle consensus features are the same feature positions that exist in the same spatiotemporal order of the target feature paths corresponding to different industrial endoscopes. The spatial coordinates, texture features, and morphological parameters of the feature positions are consistent in all target feature paths.

8. The industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction according to claim 7, characterized in that, The process of retrieving consensus feature pairs from various multi-angle perspectives, traversing the target feature path, and determining whether there are deviations in the target feature paths of various industrial endoscopes based on the traversal results includes: All multi-angle consensus features are integrated to form a consensus feature set. Based on the consensus feature set, all target feature paths are traversed and verified. During the traversal and verification process, the feature verification pointer traverses the trajectory of the target feature path point by point, and the spatial coordinates and feature parameters of each feature point are compared with the multi-angle consensus features of the corresponding spatiotemporal location in the consensus feature set. If, after traversal and verification, the deviation values ​​of all feature points on the target feature path are less than the deviation threshold, then it is determined that the industrial endoscope acquisition corresponding to the target feature path has no deviation. If the deviation value of a feature point on the target feature path is greater than or equal to the deviation threshold, it is determined that there is a sampling deviation on the target feature path, and the corresponding industrial endoscope sampling accuracy does not meet the requirements. If the spatial coordinates of the deviation feature point are offset in the horizontal direction, it is determined to be a sampling angle deviation. The verification and comparison module sends a debugging command to the angle adjustment pan-tilt unit of the industrial endoscope. Each time the debugging command is executed, the result measurement image of the corresponding area is re-acquired, and a new target feature path is generated.

9. The industrial endoscope measurement system based on binocular image acquisition and AI intelligent correction according to claim 8, characterized in that, If the number of consecutive executions of the debugging command reaches the preset threshold, and the deviation feature points on the target feature path still have deviations exceeding the deviation threshold, and the deviation value does not show an obvious decreasing trend, then it is determined that the acquisition deviation is not caused by improper acquisition parameter settings, but by the hardware defects of the industrial endoscope itself. Then, the verification and comparison module uses the spatial coordinates, deviation direction and deviation value of the deviation feature points, combined with the current acquisition angle and height of the corresponding industrial endoscope, to obtain the source location of the deviation through three-dimensional spatial inverse operation, that is, the defect location of the industrial endoscope.

Images