A method and system for detecting and identifying surface defects of a gypsum board

Abstract

The application discloses a gypsum board surface defect detection and identification method and system, which comprises the following steps: projecting flat light and structural stripe light to the board surface by using a projector, and collecting images of the gypsum board on the gypsum board conveying mechanism by using a camera to form a structural light board surface image and a flat light board surface image respectively; carrying out image processing on the collected structural light board surface image and flat light board surface image, calculating the curvature feature of the structural stripe light coding of the depth map, and extracting the edge feature of the gypsum board surface image; combining the image processing results of the depth map and the gypsum board surface image to identify the pixel position corresponding to the defects on the gypsum board surface; projecting the structural stripe light to the board surface by using the projector, learning the structural light board surface image through a deep learning model, and identifying whether the gypsum board surface has defects and the specific position of each defect; and the application can identify and determine the defects of the structural light board surface image and the position corresponding to the defects, thereby improving the accuracy of defect identification.

Description

Technical Field

[0001] This invention relates to the field of gypsum board surface defect technology, specifically to a method and system for detecting and identifying gypsum board surface defects. Background Technology

[0002] Gypsum board, a widely used material in the construction industry, directly affects the aesthetics and performance of the final product due to its surface quality. During the production process, gypsum board may encounter defects such as cracks, bubbles, scratches, grooves, low board temperature, and overheating, resulting in substandard products. Traditional inspection methods rely on manual visual inspection, which is inefficient, prone to omissions, and involves high labor intensity.

[0003] With the development of image processing technology and automation control technology, by selecting different types of sensors, it is possible to quickly and accurately detect defects on the surface of gypsum board. However, most of the existing sensor selections for defect detection on gypsum board surfaces are: laser rangefinders to detect unevenness on the surface of gypsum board, and high-resolution industrial cameras to complete image acquisition, processing and recognition, and to identify bubbles, scratches and cracks on the surface of gypsum board.

[0004] However, when existing laser rangefinders detect unevenness on the surface of gypsum board, most can only detect the flatness of a single line on the surface of the gypsum board. Therefore, if the flatness of the entire surface of the gypsum board is to be monitored, a densely distributed array of laser rangefinders is required. As a result, the demand for laser rangefinders is large, and the amount of data processing is also large. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for detecting and identifying defects on the surface of gypsum board, in order to solve the problems of existing technologies.

[0006] When monitoring the flatness of the entire surface of gypsum board, it is necessary to set up closely distributed laser rangefinders. Therefore, there is a large demand for laser rangefinders and a large amount of data processing technology.

[0007] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution:

[0008] A method for detecting and identifying defects on the surface of gypsum board includes the following steps:

[0009] Step 100: Set the camera above the gypsum board conveying mechanism, project flat light and structural stripe light onto the board surface using a projector, and use the camera to capture images of the gypsum board on the gypsum board conveying mechanism, obtain alternating images of the structured light board surface and the flat light board surface, and stitch the structured light board surface images together in the order of shooting to form a structured light board surface image, and stitch the flat light board surface images together to form a flat light board surface image.

[0010] Step 200: Perform image processing on the acquired structured light board surface image and flat light board surface image to generate a depth map containing only structured stripe light coding and a filtered and enhanced gypsum board surface image, respectively. Calculate the curvature features of the structured stripe light coding in the depth map and extract the edge features of the gypsum board surface image.

[0011] Step 300: Combining the image processing results of the depth map and the gypsum board surface image, identify the pixel positions corresponding to cracks, bubbles, scratches, and surface flatness defects on the gypsum board surface.

[0012] As a preferred embodiment of the present invention, in step 100, the number of shooting groups of the camera is adjusted based on the transmission speed of the gypsum board transmission mechanism, so that there are fixed overlapping pixels in the images captured by two adjacent groups of the camera.

[0013] In each set of images captured by the camera, the structured light projector is first turned off to capture a flat light panel image, and then the structured light projector is turned on to generate continuous striped structured light on the surface of the gypsum board to capture a structured light panel image.

[0014] The structured light panel images captured in sequence are stitched together to form a structured light panel image of the entire gypsum board surface, and the flat light panel images captured in sequence are stitched together to form a flat light panel image of the entire gypsum board surface.

[0015] The structured light panel image of the entire gypsum board surface formed by splicing is cut, and the flat light panel image of the entire gypsum board surface formed by splicing is also cut, retaining only the information containing the gypsum board surface.

[0016] As a preferred embodiment of the present invention, a detection unit is provided upstream of the camera. The detection unit is used to monitor the plasterboard that is about to be transmitted to the camera, and to control the camera to start shooting and stop shooting based on the monitoring results of the detection unit.

[0017] The structured light panel images and flat light panel images, taken sequentially according to the number of shooting groups, are numbered in order.

[0018] Assuming there are n groups of images, select 1, 3, 5, 7, ..., 2n-1 flat light plate surface images and stitch them together in sequence to form a flat light plate surface image;

[0019] Select 2, 4, 6, 8, ..., 2n structured light panel images and stitch them together sequentially to form a structured light panel image.

[0020] As a preferred embodiment of the present invention, in step 100, before acquiring the alternating captured structured light plate surface images and planar light plate surface images, the camera's intrinsic and extrinsic parameters are calibrated. The specific calibration method is as follows:

[0021] Using a checkerboard pattern on a plasterboard conveyor, the camera captures multiple images of the checkerboard at different locations.

[0022] The corresponding projected pixel coordinates are obtained through projection / decoding, and projector intrinsic parameter estimation is performed on these points.

[0023] Custom optimization simultaneously solves for the intrinsic and extrinsic parameters of both the camera and the projector;

[0024] Output camera intrinsic parameters, projector intrinsic parameters, camera extrinsic parameters, projector intrinsic parameters, and distortion parameters to calibrate the projector and camera respectively.

[0025] As a preferred embodiment of the present invention, in step 200, the image processing of the acquired structured light plate image and the flat light plate image is performed as follows:

[0026] The stitched structured light panel image and the flat light panel image are divided into multiple structured light segmented images and flat light segmented images according to the same scale.

[0027] Phase calculation and decoding, phase unwrapping and denoising processing are performed on the stripe sequence of the structured light segmented image to restore the depth map;

[0028] The flat, light-segmented image is subjected to bilateral filtering / guided filtering for noise reduction while preserving the edges. Local contrast is enhanced by adaptive histogram equalization to retain defects on the gypsum board surface, thus obtaining the gypsum board surface image.

[0029] Two depth maps and a gypsum board surface image of the same scale are selected. Based on the curvature features of the depth map, the pixels corresponding to the curvature changes of the structured light are identified. The pixels corresponding to defects are identified from the gypsum board surface image, so as to identify cracks, bubbles, scratches and surface flatness defects on the gypsum board surface.

[0030] As a preferred embodiment of the present invention, the method for generating a depth map containing only structural stripe optical coding is as follows:

[0031] Phase shifting phase calculation: Calculate the phase value of each pixel in the structured light segmented image to obtain the packaging phase;

[0032] Phase expansion: Obtain integer period indices and synthesize absolute phases;

[0033] The disparity / depth of the structured light segmented image is determined based on phase: the phase of each pixel is mapped to the projected pixel through geometric relationships, and then the depth is obtained using triangulation, specifically:

[0034] By using the relationship between phase and column index, the projected pixel corresponding to the camera pixel can be found;

[0035] Use the camera pixel coordinates, projected pixel coordinates, and their projection matrices to perform triangulation to find 3D points.

[0036] As a preferred embodiment of the present invention, the method for calculating the curvature features of the structural stripe optical encoding of the depth map is as follows:

[0037] The depth map is denoised using median / bilateral filtering, and occlusion holes are filled using multi-view merging.

[0038] The local gradient of the depth map is used to calculate the normal (Nx, Ny, Nz), and the second derivative or principal curvature is used to estimate the local curvature. Cracks and scratches on the plasterboard are usually represented by high curvature or abrupt normal bands, bubbles and bulges on the plasterboard are represented by low-frequency high protrusions, and unevenness on the plasterboard is also represented by low-frequency high protrusions.

[0039] A two-dimensional coordinate system is constructed for each depth map to determine the pixel coordinate values ​​corresponding to the pixels with curvature changes.

[0040] As a preferred embodiment of the present invention, the method for extracting the edge features of the gypsum board surface image is as follows:

[0041] The bright or dark areas of the gypsum board surface image are captured using a morphological cap, and the location of the defect is screened to determine whether it is an air bubble based on the roundness / aspect ratio of the bright or dark areas.

[0042] Set a pixel threshold, filter out abnormal pixels by comparing the results of the pixel threshold, perform connected component analysis on the abnormal pixels, and identify scratches or cracks by the aspect ratio of the structure formed by the connected components.

[0043] A two-dimensional coordinate system is constructed for each of the gypsum board surface images to determine the pixel coordinate values ​​corresponding to pixels with pixel anomalies.

[0044] As a preferred embodiment of the present invention, in step 300, the pixel coordinate values ​​of curvature changes identified by the curvature features of the structural stripe optical encoding based on the depth map are compared with the pixel coordinate values ​​corresponding to the pixel points with pixel anomalies determined based on the gypsum board surface image, so as to distinguish the defect coordinate values ​​in the depth map caused by the unevenness of the gypsum board surface, the defect coordinate values ​​caused by the air bubbles on the gypsum board surface, and the defect coordinate values ​​caused by the scratches and cracks on the gypsum board surface.

[0045] Deep learning was performed to acquire the structured light curvature features corresponding to the coordinate values ​​of defects caused by unevenness of the gypsum board surface, the structured light curvature features corresponding to the coordinate values ​​of defects caused by bubbles on the gypsum board surface, and the structured light curvature features corresponding to the coordinate values ​​of defects caused by scratches and cracks on the gypsum board surface, so that the unevenness, bubbles and scratches and cracks on the gypsum board surface can be identified based on the structured light curvature features.

[0046] The structured fringe light is projected onto the board surface using a projector, and the structured light board surface segment is captured by a camera. After the captured structured light board surface segment is stitched together to form a structured light board surface image, the curvature characteristics of the structured fringe light encoding of the structured light board surface image are calculated.

[0047] Based on curvature characteristics, defects such as unevenness, bubbles, scratches, and cracks on the surface of gypsum board are identified.

[0048] In addition, the present invention also provides an identification system for a method of detecting and identifying defects on the surface of gypsum board, comprising:

[0049] A photoelectric monitoring unit is installed on the gypsum board conveying mechanism, the photoelectric monitoring unit being used to identify the gypsum board being conveyed to the photoelectric monitoring unit;

[0050] The camera located downstream of the photoelectric monitoring unit is capable of projecting flat light and structural stripe light sequentially onto the surface of the gypsum board through a projector, and can also capture images of the gypsum board surface through the camera.

[0051] A control system is connected to the photoelectric monitoring unit and the camera, and the control system adjusts the start-up and shutdown of the camera based on the monitoring results of the photoelectric monitoring unit;

[0052] The structured light image processing module is used to stitch together images containing structured stripe light sequentially, convert the images containing structured stripe light into depth maps, calculate the curvature features of the structured stripe light encoding in the depth map, and identify defects on the gypsum board surface based on the curvature features.

[0053] The flat light image processing module is used to stitch together images containing flat light in sequence, convert the images containing flat light into gypsum board surface images, extract the edge features of the gypsum board surface images, and identify defects on the gypsum board surface based on the edge features.

[0054] The defect differentiation module is used to perform pixel-level mapping between the defects on the gypsum board surface identified by the structured light image processing module and the defects on the gypsum board surface identified by the smooth light image processing module, so as to determine the pixel positions corresponding to the unevenness, bubbles and scratches / cracks of the gypsum board surface respectively.

[0055] Compared with the prior art, the present invention has the following advantages:

[0056] This invention first stitches together the structured light panel surface images to form a structured light panel surface image, and then stitches together the flat light panel surface images to form a flat light panel surface image. The defect coordinate values ​​identified based on the stripe deformation curvature of the structured light panel surface image are compared with the defect coordinate values ​​identified based on the flat light panel surface image processing. This allows for the determination of the pixel positions corresponding to cracks, bubbles, scratches, and surface flatness defects in the structured light panel surface image.

[0057] Then, deep learning is performed on the curvature features of structured light corresponding to cracks, bubbles, scratches, and surface flatness defects. When the structured light image of the board surface is reprocessed, based on the deep learning results, the cracks, bubbles, scratches, and surface flatness defects in the structured light image of the board surface and the location of the defects can be identified. Thus, the accuracy of defect identification is improved through verification. Attached Figure Description

[0058] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0059] Figure 1 This is a flowchart illustrating the method for detecting and identifying surface defects in gypsum board according to an embodiment of the present invention.

[0060] Figure 2 This is a schematic diagram of the gypsum board surface defect detection and identification system according to an embodiment of the present invention;

[0061] The labels in the diagram represent the following:

[0062] 1-Photoelectric monitoring unit; 2-Camera; 3-Control system; 4-Structured light image processing module; 5-Plain light image processing module; 6-Defect differentiation module. Detailed Implementation

[0063] 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, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] like Figure 1As shown, the present invention provides a method for detecting and identifying defects on the surface of gypsum board, comprising the following steps:

[0065] Step 100: Set the camera above the gypsum board conveying mechanism, project flat light and structural stripe light onto the board surface using a projector, and use the camera to capture images of the gypsum board on the gypsum board conveying mechanism. Obtain alternating images of the structured light board surface and the flat light board surface, and stitch the structured light board surface images together in the order of shooting to form a structured light board surface image. Then, stitch the flat light board surface images together to form a flat light board surface image.

[0066] Step 200: Perform image processing on the acquired structured light plate surface image and flat light plate surface image to generate a depth map containing only the structured stripe light coding and a filtered and enhanced gypsum board surface image, respectively. Calculate the curvature features of the structured stripe light coding in the depth map and extract the edge features of the gypsum board surface image.

[0067] Step 300: Combine the image processing results of the depth map and the gypsum board surface image to identify the pixel positions corresponding to cracks, bubbles, scratches and surface flatness defects on the gypsum board surface, and perform deep learning on the curvature features corresponding to cracks, bubbles, scratches and surface flatness defects on the gypsum board surface.

[0068] Step 400: Project structural stripe light onto the board surface using a projector. Use a deep learning model to learn the structured light board surface image and identify whether there are cracks, bubbles, scratches, and surface flatness defects on the gypsum board surface, as well as the specific location of each defect.

[0069] Because "paper-faced gypsum board" is usually without obvious texture and has a relatively smooth surface, this embodiment projects known stripes or phase codes onto the surface of the gypsum board using a projector. When defects appear on the surface of the gypsum board, the stripes in the image captured by the camera are deformed, and the defects are identified by calculating the curvature of the stripe deformation.

[0070] After turning off the structural stripe light and retaining only the flat light, the camera takes an image of the gypsum board surface with only the flat light as the light source. By performing image processing on the gypsum board surface image and extracting the edge features of the gypsum board surface image, cracks, bubbles, scratches and other defects on the gypsum board surface can be identified.

[0071] Since the deformation curvature of the stripes alone cannot accurately distinguish cracks, bubbles, scratches, and surface flatness defects on the gypsum board surface, this application sequentially stitches together the structured light board surface images to form a structured light board surface image, and then stitches together the flat light board surface images to form a flat light board surface image. The defect coordinate values ​​identified based on the deformation curvature of the stripes are compared with the defect coordinate values ​​identified based on the flat light board surface image processing. This allows the pixel positions corresponding to cracks, bubbles, scratches, and surface flatness defects on the gypsum board surface to be determined separately.

[0072] The working principle of identifying defects by stripe deformation curvature is as follows:

[0073] Structured light relies on a clear, recognizable projection of a pattern onto a surface. Any alteration to surface reflection / geometry (abrupt reflectivity changes, scattering, occlusion) will affect the encoding / decoding of fringes and phase recovery. Light scattering in thin layers, bubbles, or multiple reflections on crack sidewalls can cause geometric distortion or blurring of the projection pattern, leading to geometric reconstruction errors.

[0074] In step 100, the number of camera shooting groups is adjusted based on the transmission speed of the gypsum board transmission mechanism, so that fixed pixels overlap in the images captured by adjacent groups of cameras.

[0075] When the camera captures each set of images, first turn off the structured light projector to capture a flat light panel image, then turn on the structured light projector to generate continuous striped structured light on the surface of the gypsum board, and capture a structured light panel image.

[0076] The structured light panel images captured in sequence are stitched together to form a complete structured light panel image of the gypsum board surface. Similarly, the flat light panel images captured in sequence are stitched together to form a complete flat light panel image of the gypsum board surface.

[0077] The structured light panel image of the entire gypsum board surface formed by splicing is cut, and the flat light panel image of the entire gypsum board surface formed by splicing is also cut, retaining only the information containing the gypsum board surface.

[0078] It should be noted that in step 100 of this embodiment, two images can be captured sequentially each time. One image uses flat light to capture a flat light panel image, which is specifically an image of the gypsum board surface. The other image uses structured stripe light to capture a structured light panel image, which is specifically an image of the structured stripe light on the gypsum board surface.

[0079] The flat light plate images output from two adjacent shooting sessions can be stitched together, and similarly, the structured stripe light images output from two adjacent shooting sessions can also be stitched together.

[0080] A detection unit is located upstream of the camera. The detection unit is used to monitor the plasterboard that is about to be transmitted to the camera. Based on the monitoring results of the detection unit, the camera is controlled to start shooting and stop shooting respectively.

[0081] The structured light panel images and flat light panel images, taken sequentially according to the number of shooting groups, are numbered in order.

[0082] Assuming there are n groups of images, select 1, 3, 5, 7, ..., 2n-1 flat light plate surface images and stitch them together in sequence to form a flat light plate surface image;

[0083] Select 2, 4, 6, 8, ..., 2n structured light panel images and stitch them together sequentially to form a structured light panel image.

[0084] After sequentially stitching together all flat light plate surface images to form a flat light plate surface image, and sequentially stitching together all structured light plate surface images to form a structured light plate surface image, image processing is performed on the acquired structured light plate surface images to generate depth maps containing only structured stripe optical codes. The overall process for calculating the curvature features of the structured stripe optical codes in the depth maps is as follows:

[0085] Acquisition: A structured light sequence (Gray code + phase shift) is projected using a projector, and multiple frames are captured by a camera.

[0086] Calibration: Calibrate the camera and projector (treating the projector as an inverse camera) to obtain the camera's intrinsic parameters, the projector's intrinsic parameters, and the extrinsic parameters (baseline) of both.

[0087] Phase calculation and depth reconstruction: The phase of the phase-shifted sequence is calculated, the phase is unfolded using Gray code, and the depth / height map is obtained by combining camera-projection geometric triangulation.

[0088] Preprocessing: Denoising, hole repair, calculation of normals and curvature, generation of phase confidence / residual map.

[0089] Before acquiring alternating images of the structured light and flat light plates, the camera's intrinsic and extrinsic parameters are calibrated. The specific calibration method is as follows:

[0090] Using a checkerboard pattern on a plasterboard conveyor, multiple images of the checkerboard at different locations are captured by a camera.

[0091] The corresponding projected pixel coordinates are obtained through projection / decoding, and projector intrinsic parameter estimation is performed on these points.

[0092] Custom optimization simultaneously solves for the intrinsic and extrinsic parameters of both the camera and the projector;

[0093] Output camera intrinsic parameters, projector intrinsic parameters, camera extrinsic parameters, projector intrinsic parameters, and distortion parameters to calibrate the projector and camera respectively.

[0094] The purpose of camera intrinsic and extrinsic calibration can be simply summarized as: to accurately link pixel coordinates with 3D world coordinates, eliminate lens distortion and uncertainties in camera position / orientation, thereby enabling quantitative geometric calculations (range measurement, angle measurement, reconstruction, registration, projection, etc.).

[0095] In step 200, the image processing of the acquired structured light panel image and flat light panel image is performed as follows:

[0096] The stitched structured light panel image and the flat light panel image are divided into multiple structured light segmented images and flat light segmented images according to the same scale.

[0097] Phase calculation and decoding, phase unwrapping and denoising are performed on the stripe sequence of the structured light segmented image to restore the depth map.

[0098] The flat, segmented image is denoised by bilateral filtering / guided filtering while preserving the edges. Local contrast is enhanced by adaptive histogram equalization to retain defects on the gypsum board surface, thus obtaining the gypsum board surface image.

[0099] Two depth maps and gypsum board surface images of the same scale are selected. Based on the curvature features of the depth map, the pixels corresponding to the curvature changes of the structured light are identified. The pixels corresponding to defects are identified from the gypsum board surface images to identify cracks, bubbles, scratches and surface flatness defects on the gypsum board surface.

[0100] The method for generating a depth map containing only structural stripe optical coding is as follows:

[0101] Phase shift solution: Calculate the phase value of each pixel in the structured light segmented image to obtain the packaging phase.

[0102] Phase unrolling: Obtain integer period indices and synthesize absolute phases.

[0103] Disparity / depth of structured light segmented images based on phase determination: The phase of each pixel is mapped to the projected pixel through geometric relationships, and the depth is obtained using triangulation, specifically as follows:

[0104] By using the relationship between phase and column index, the projected pixel corresponding to the camera pixel can be found;

[0105] Use the camera pixel coordinates, projected pixel coordinates, and their projection matrices to perform triangulation to find 3D points.

[0106] In this embodiment, since the structured light panel image is specifically a structured stripe light image of the gypsum board surface, if there are cracks, bubbles, scratches, or surface flatness defects on the gypsum board surface, the structured stripe light in the structured stripe light image will bend at the defect location. The flatness panel image is specifically a gypsum board surface image. By comparing pixel values, defects such as bubbles, cracks, and scratches on the gypsum board surface can be identified.

[0107] The method for calculating the curvature features of the structural stripe optical encoding in the depth map is as follows:

[0108] Median / bilateral filtering is applied to the depth map for noise reduction, and occlusion holes are filled using multi-view merging.

[0109] The local gradient of the depth map is used to calculate the normal (Nx, Ny, Nz), and then the second derivative or principal curvature is used to estimate the local curvature.

[0110] For each depth map, construct a two-dimensional coordinate system and determine the pixel coordinate values ​​corresponding to the pixels with curvature changes.

[0111] By performing image processing on the depth map, the curvature features of the structural stripe optical encoding of the depth map are calculated. Cracks and scratches on the gypsum board usually appear as high curvature or abrupt normal bands, while bubbles and bulges on the gypsum board appear as low-frequency high-protrusions. Unevenness on the gypsum board also appears as low-frequency high-protrusions. Therefore, when there are changes in curvature features in the depth map, it indicates that there are some defects on the surface of the gypsum board.

[0112] The method for extracting edge features from a gypsum board surface image is as follows:

[0113] Use a morphological cap to capture the bright or dark areas of the gypsum board surface image, and filter whether the defect location is an air bubble by the roundness / aspect ratio of the bright or dark area.

[0114] Set a pixel threshold, filter out abnormal pixels by comparing the results of the pixel threshold, perform connected component analysis on the abnormal pixels, and identify scratches or cracks by the aspect ratio of the structure formed by the connected components.

[0115] A two-dimensional coordinate system is constructed for each gypsum board surface image to determine the pixel coordinate values ​​corresponding to pixels with pixel anomalies.

[0116] When a flat light is used to illuminate the surface of a gypsum board, defects such as bubbles, scratches, or cracks can be identified by comparing pixel values. After identifying the defects on the gypsum board surface using two different methods, it should be noted that the presence of curvature features in the depth map indicates the presence of some defects on the gypsum board surface. However, it is difficult to determine the specific defects based on the curvature features of structured light. By extracting the edge features of the gypsum board surface image under flat light, bubble defects, scratches, and cracks can be specifically identified, and the corresponding pixel coordinate values ​​can be determined.

[0117] By comparing the pixel coordinates of defects identified based on the curvature features of structured light with the pixel coordinates of specific defects determined by the curvature features of structured light, the defects identified by the curvature features of structured light can be specifically classified into bubble defects, scratch and crack defects, and board surface flatness defects.

[0118] In step 300, the pixel coordinates of the curvature changes identified by the curvature features of the structural stripe optical encoding based on the depth map are compared with the pixel coordinates of the pixel points corresponding to the pixel anomalies determined based on the gypsum board surface image. This distinguishes the defect coordinates caused by the unevenness of the gypsum board surface, the defect coordinates caused by air bubbles on the gypsum board surface, and the defect coordinates caused by scratches and cracks on the gypsum board surface in the depth map.

[0119] Deep learning was performed to obtain the structured light curvature features corresponding to the coordinate values ​​of defects caused by unevenness of the gypsum board surface, the structured light curvature features corresponding to the coordinate values ​​of defects caused by bubbles on the gypsum board surface, and the structured light curvature features corresponding to the coordinate values ​​of defects caused by scratches and cracks on the gypsum board surface, so that the unevenness, bubbles and scratches and cracks on the gypsum board surface can be identified based on the structured light curvature features.

[0120] Structured fringe light is projected onto the board surface using a projector, and the structured light board surface segment is captured using a camera. After the captured structured light board surface segment is stitched together to form a structured light board surface image, the curvature characteristics of the structured fringe light encoding in the structured light board surface image are calculated.

[0121] Based on curvature characteristics, defects such as unevenness, bubbles, scratches, and cracks on the surface of gypsum board are identified.

[0122] In addition, such as Figure 2 As shown, the present invention also provides an identification system for a method of detecting and identifying defects on the surface of gypsum board, comprising: a photoelectric monitoring unit 1 disposed on a gypsum board transmission mechanism, the photoelectric monitoring unit 1 being used to identify the gypsum board transmitted to the photoelectric monitoring unit.

[0123] The structured light camera 2 is located downstream of the photoelectric monitoring unit. The structured light camera 2 can project flat light and structured stripe light onto the surface of the gypsum board in sequence through a projector, and capture images of the surface of the gypsum board through the camera.

[0124] The control system 3 is connected to the photoelectric monitoring unit 1 and the structured light camera 2. The control system 3 adjusts the start-up and shutdown of the structured light camera 2 based on the monitoring results of the photoelectric monitoring unit 1.

[0125] The structured light image processing module 4 is used to stitch together the captured images containing structured stripe light sequentially, convert the captured images containing structured stripe light into depth maps, calculate the curvature features of the structured stripe light encoding in the depth maps, and identify defects on the gypsum board surface based on the curvature features.

[0126] The flat light image processing module 5 is used to stitch together the captured images containing flat light in sequence, convert the captured images containing flat light into gypsum board surface images, extract the edge features of the gypsum board surface images, and identify defects on the gypsum board surface based on the edge features.

[0127] The defect differentiation module 6 is used to perform pixel-level mapping between the defects on the gypsum board surface identified by the structured light image processing module 4 and the defects on the gypsum board surface identified by the smooth light image processing module 5, so as to determine the pixel positions corresponding to the unevenness, bubbles and scratches / cracks of the gypsum board surface respectively.

[0128] In this embodiment, the structured light panel surface images are first stitched together to form a structured light panel surface image. After the flat light panel surface images are stitched together to form a flat light panel surface image, the defect coordinate values ​​identified based on the stripe deformation curvature of the structured light panel surface image are compared with the defect coordinate values ​​identified based on the flat light panel surface image processing. This allows the pixel positions corresponding to cracks, bubbles, scratches, and surface flatness defects in the structured light panel surface image to be determined respectively.

[0129] Then, deep learning is performed on the curvature features of structured light corresponding to cracks, bubbles, scratches, and surface flatness defects. When the structured light image of the board surface is reprocessed, based on the deep learning results, the cracks, bubbles, scratches, and surface flatness defects in the structured light image of the board surface and the location of the defects can be identified. Thus, the accuracy of defect identification is improved through verification.

[0130] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.

Claims

1. A method for detecting and identifying defects on the surface of gypsum board, characterized in that, Includes the following steps: Step 100: Set the camera above the gypsum board conveying mechanism, project flat light and structural stripe light onto the board surface using a projector, and use the camera to capture images of the gypsum board on the gypsum board conveying mechanism, obtain alternating images of the structured light board surface and the flat light board surface, and stitch the structured light board surface images together in the order of shooting to form a structured light board surface image, and stitch the flat light board surface images together to form a flat light board surface image. Step 200: Perform image processing on the acquired structured light board surface image and flat light board surface image to generate a depth map containing only structured stripe light coding and a filtered and enhanced gypsum board surface image, respectively. Calculate the curvature features of the structured stripe light coding in the depth map and extract the edge features of the gypsum board surface image. Step 300: Combining the image processing results of the depth map and the gypsum board surface image, identify the pixel positions corresponding to cracks, bubbles, scratches and surface flatness defects on the gypsum board surface. Step 400: Project structural stripe light onto the board surface using a projector. Use a deep learning model to learn the structured light board surface image and identify whether there are cracks, bubbles, scratches, and surface flatness defects on the gypsum board surface, as well as the specific location of each defect.

2. The method for detecting and identifying surface defects in gypsum board according to claim 1, characterized in that, In step 100, the number of camera shooting groups is adjusted based on the transmission speed of the gypsum board transmission mechanism, so that a fixed number of pixels overlap in the images captured by two adjacent groups of the camera. In each set of images captured by the camera, the structured light projector is first turned off to capture a flat light panel image, and then the structured light projector is turned on to generate continuous striped structured light on the surface of the gypsum board to capture a structured light panel image. The structured light panel images captured in sequence are stitched together to form a structured light panel image of the entire gypsum board surface, and the flat light panel images captured in sequence are stitched together to form a flat light panel image of the entire gypsum board surface. The structured light panel image of the entire gypsum board surface formed by splicing is cut, and the flat light panel image of the entire gypsum board surface formed by splicing is also cut, retaining only the information containing the gypsum board surface.

3. The method for detecting and identifying surface defects in gypsum board according to claim 2, characterized in that, A detection unit is provided upstream of the camera. The detection unit is used to monitor the plasterboard that is about to be transmitted to the camera. Based on the monitoring results of the detection unit, the camera is controlled to start and stop shooting respectively. The structured light panel images and flat light panel images, taken sequentially according to the number of shooting groups, are numbered in order. Assuming there are n groups of images, select 1, 3, 5, 7, ..., 2n-1 flat light plate surface images and stitch them together in sequence to form a flat light plate surface image; Select 2, 4, 6, 8, ..., 2n structured light panel images and stitch them together sequentially to form a structured light panel image.

4. The method for detecting and identifying surface defects in gypsum board according to claim 1, characterized in that, In step 100, before acquiring the alternating images of the structured light plate and the planar light plate, the camera's intrinsic and extrinsic parameters are calibrated. The specific calibration method is as follows: Using a checkerboard pattern on a plasterboard conveyor, the camera captures multiple images of the checkerboard at different locations. The corresponding projected pixel coordinates are obtained through projection / decoding, and projector intrinsic parameter estimation is performed on these points. Custom optimization simultaneously solves for the intrinsic and extrinsic parameters of both the camera and the projector; Output camera intrinsic parameters, projector intrinsic parameters, camera extrinsic parameters, projector intrinsic parameters, and distortion parameters to calibrate the projector and camera respectively.

5. The method for detecting and identifying surface defects in gypsum board according to claim 1, characterized in that, In step 200, the image processing of the acquired structured light panel image and flat light panel image is performed as follows: The stitched structured light panel image and the flat light panel image are divided into multiple structured light segmented images and flat light segmented images according to the same scale. Phase calculation and decoding, phase unwrapping and denoising processing are performed on the stripe sequence of the structured light segmented image to restore the depth map; The flat, light-segmented image is subjected to bilateral filtering / guided filtering for noise reduction while preserving the edges. Local contrast is enhanced by adaptive histogram equalization to retain defects on the gypsum board surface, thus obtaining the gypsum board surface image. Two depth maps and a gypsum board surface image of the same scale are selected. Based on the curvature features of the depth map, the pixels corresponding to the curvature changes of the structured light are identified. The pixels corresponding to defects are identified from the gypsum board surface image, so as to identify cracks, bubbles, scratches and surface flatness defects on the gypsum board surface.

6. The method for detecting and identifying surface defects in gypsum board according to claim 5, characterized in that, The method for generating a depth map containing only structural stripe optical coding is as follows: Phase shifting phase calculation: Calculate the phase value of each pixel in the structured light segmented image to obtain the packaging phase; Phase expansion: Obtain integer period indices and synthesize absolute phases; The disparity / depth of the structured light segmented image is determined based on phase: the phase of each pixel is mapped to the projected pixel through geometric relationships, and then the depth is obtained using triangulation, specifically: By using the relationship between phase and column index, the projected pixel corresponding to the camera pixel can be found; Use the camera pixel coordinates, projected pixel coordinates, and their projection matrices to perform triangulation to find 3D points.

7. The method for detecting and identifying surface defects in gypsum board according to claim 6, characterized in that, The method for calculating the curvature features of the optically encoded structural stripes of the depth map is as follows: The depth map is denoised using median / bilateral filtering, and occlusion holes are filled using multi-view merging. The local gradient of the depth map is used to calculate the normal (Nx, Ny, Nz), and the second derivative or principal curvature is used to estimate the local curvature. Cracks and scratches on the plasterboard are usually represented by high curvature or abrupt normal bands, bubbles and bulges on the plasterboard are represented by low-frequency high protrusions, and unevenness on the plasterboard is also represented by low-frequency high protrusions. A two-dimensional coordinate system is constructed for each depth map to determine the pixel coordinate values ​​corresponding to the pixels with curvature changes.

8. The method for detecting and identifying surface defects in gypsum board according to claim 7, characterized in that, The method for extracting edge features from the gypsum board surface image is as follows: The bright or dark areas of the gypsum board surface image are captured using a morphological cap, and the location of the defect is screened to determine whether it is an air bubble based on the roundness / aspect ratio of the bright or dark areas. Set a pixel threshold, filter out abnormal pixels by comparing the results of the pixel threshold, perform connected component analysis on the abnormal pixels, and identify scratches or cracks by the aspect ratio of the structure formed by the connected components. A two-dimensional coordinate system is constructed for each of the gypsum board surface images to determine the pixel coordinate values ​​corresponding to pixels with pixel anomalies.

9. The method for detecting and identifying surface defects in gypsum board according to claim 8, characterized in that, In step 300, the pixel coordinate values ​​of curvature changes identified by the curvature features of the structural stripe optical encoding based on the depth map are compared with the pixel coordinate values ​​corresponding to the pixel points with pixel anomalies determined based on the gypsum board surface image, so as to distinguish the defect coordinate values ​​in the depth map caused by the unevenness of the gypsum board surface, the defect coordinate values ​​caused by the air bubbles on the gypsum board surface, and the defect coordinate values ​​caused by the scratches and cracks on the gypsum board surface. Deep learning was performed to acquire the structured light curvature features corresponding to the coordinate values ​​of defects caused by unevenness of the gypsum board surface, the structured light curvature features corresponding to the coordinate values ​​of defects caused by bubbles on the gypsum board surface, and the structured light curvature features corresponding to the coordinate values ​​of defects caused by scratches and cracks on the gypsum board surface, so that the unevenness, bubbles and scratches and cracks on the gypsum board surface can be identified based on the structured light curvature features. The structured fringe light is projected onto the board surface using a projector, and the structured light board surface segment is captured by a camera. After the captured structured light board surface segment is stitched together to form a structured light board surface image, the curvature characteristics of the structured fringe light encoding of the structured light board surface image are calculated. Based on curvature characteristics, defects such as unevenness, bubbles, scratches, and cracks on the surface of gypsum board are identified.

10. An identification system based on the gypsum board surface defect detection and identification method according to any one of claims 1-9, characterized in that, include: A photoelectric monitoring unit (1) is installed on the gypsum board transmission mechanism. The photoelectric monitoring unit (1) is used to identify the gypsum board transmitted to the photoelectric monitoring unit. The camera (2) located downstream of the photoelectric monitoring unit is capable of projecting flat light and structural stripe light onto the surface of the gypsum board in sequence through a projector, and capturing images of the surface of the gypsum board through the camera. The control system (3) is connected to the photoelectric monitoring unit (1) and the camera (2). The control system (3) adjusts the start-up and shutdown of the camera (2) based on the monitoring results of the photoelectric monitoring unit (1). The structured light image processing module (4) is used to stitch together the captured images containing structured stripe light in sequence, convert the captured images containing structured stripe light into a depth map, and calculate the curvature features of the structured stripe light encoding of the depth map, and identify the defects on the surface of the gypsum board based on the curvature features. The flat light image processing module (5) is used to stitch together the captured images containing flat light in sequence, convert the captured images containing flat light into gypsum board surface images, extract the edge features of the gypsum board surface images, and identify the defects on the gypsum board surface based on the edge features. The defect differentiation module (6) is used to perform pixel-level correspondence between the defects on the gypsum board surface identified by the structured light image processing module (4) and the defects on the gypsum board surface identified by the smooth light image processing module (5), so as to determine the pixel positions corresponding to the unevenness, bubbles and scratches / cracks of the gypsum board surface respectively.

Images