An image recognition-based hot-rolled strip end geometry detection system and method

Abstract

The present application belongs to the field of modern equipment manufacturing industry, and discloses a hot-rolled strip steel end geometry detection system and method based on image recognition, which comprises a fixed support module, a PLC control system, an image sensing and collecting module, a roller speed control system, a cooling system and a data collection and classification interface. Meanwhile, the CMOS linear array sensor is used for nonlinear stage measurement of the plate. The CMOS linear array sensor Draréni dynamic calibration and swarm intelligence optimization algorithm are used for parameter calibration, so as to improve the calibration accuracy and efficiency. In the detection principle method, the stage and regional detection are proposed, which can more accurately ensure the detection accuracy. Finally, the collected images are processed through the image recognition algorithm, the image stitching technology is used to complete the contour shape recognition of the high-temperature steel plate head and tail, and the contour shape and size parameters of the high-temperature plate are accurately obtained through the above method.

Description

Technical Field

[0001] This invention belongs to the field of modern equipment manufacturing industry, and specifically relates to a system and method for detecting the end geometry of hot-rolled strip steel based on image recognition. Background Technology

[0002] Steel is a crucial raw material in industrial production, playing a vital role in national economic development and serving as a foundational industry for the national economy. The head and tail shape of hot-rolled strip plays a critical role in ensuring its stability during the finishing mill rolling process. A well-designed slab head and tail shape not only effectively facilitates smooth biting and threading in the finishing mill but also profoundly impacts the successful coiling operation. After roughing, the head and tail of hot-rolled strip often exhibit geometric defects such as tongue-shaped or fishtail-shaped defects, and these areas generally have lower temperatures. Therefore, to ensure the stability of subsequent production processes, precise cutting of the head and tail portions is essential before the strip enters the finishing mill. The steel plate's profile is a vital parameter for describing its shape and straightness. Profile inspection of the steel plate's head and tail, statistical analysis of a large amount of profile inspection data, and the determination of the steel plate shearing process based on the analysis results are crucial.

[0003] Currently, manual inspection methods for medium and heavy plates are outdated and completely inadequate for inspecting the beginning and end of steel plates. Current methods for steel plate contour inspection are mainly divided into contact and non-contact methods. Contact-based plate shape inspection offers high accuracy but is expensive and suffers from wear, which reduces the accuracy of medium and heavy steel plate inspections. Non-contact plate shape inspection includes displacement methods, optical imaging methods, electromagnetic methods, and vibration measurement methods. Machine vision uses optical devices and non-contact sensors to receive images of an object, then uses computers combined with well-developed algorithms to process and calculate the images. Machine vision technology is increasingly being widely used in industrial inspection. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a system and method for detecting the end geometry of hot-rolled strip steel based on image recognition. This system can effectively improve the speed of detecting the head and tail contours of steel plates in the factory, provide rapid feedback to the shearing equipment, and improve the factory's production efficiency.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A hot-rolled strip end geometry detection system based on image recognition, the system comprising: a fixed support module, a PLC control system, an image sensor acquisition module, a roller speed control system, a cooling system, and a data acquisition and classification interface;

[0007] The fixed support module is used to support the image sensing and acquisition module;

[0008] The image sensing and acquisition module is used to acquire images of the calibration plate and the high-temperature steel plate;

[0009] The data collection and classification interface is used to classify the collected data;

[0010] The PLC control system is used for power supply and control of the entire detection system, and is responsible for signal transmission of the feedback system.

[0011] The cooling system is used to cool the detector.

[0012] The roller conveyor speed control system is used to feed back the detected target speed and the actual operating speed of the roller conveyor to the PLC control system.

[0013] Preferably, the fixed support module includes: 8 vertical camera brackets, 4 short crossbeams, 1 2-meter main crossbeam, 2 1.2-meter secondary crossbeams, 8 anti-slip pads, and 28 fixed angle irons;

[0014] The vertical camera brackets are connected by short crossbeams, which are fixed with angle irons. The short crossbeams are connected to the secondary crossbeams by angle irons, and the secondary crossbeams are connected to the main crossbeams by angle irons and bolts. The cameras are mounted on the main crossbeams by camera mounting blocks and shock-absorbing rubber pads. The vertical camera brackets are in direct contact with the ground by anti-slip feet.

[0015] Preferably, the PLC control system includes: a host computer, a communication line, a data acquisition card, a roller speed feedback module, a camera acquisition command transmission module, and an image information data transmission module.

[0016] Preferably, the image sensing and acquisition module includes: a first CMOS linear array sensor, a second CMOS linear array sensor, a first laser emitter, a second laser emitter, a first strip light source, a second strip light source, a camera fixing block, anti-slip pads, a three-axis industrial camera gimbal, and a camera image stabilization stabilizer.

[0017] The first and second CMOS linear array sensors are mounted on a three-axis industrial camera gimbal. The three-axis industrial camera gimbal is mounted on the main beam via a camera mounting block. The first and second strip light sources are mounted on a support base via light source clamps and support frames. A laser level measuring instrument is used to ensure that all CMOS linear array sensors and all light sources are on the same horizontal plane. The offset angle of the light source clamps is adjusted to ensure that the illumination area is within the target detection area. The first and second laser emitters are mounted on the main beam via camera mounting blocks to ensure that the camera detects the edge contour dimensions of the target object.

[0018] Preferably, the cooling system includes: water-cooled pipes and compressed air curtain.

[0019] Preferably, the roller conveyor speed control system includes: rollers, sprockets and chains, transmission chains, 200W servo motors, intelligent digital display speed controllers, diffuse reflection photoelectric switches, encoders, encoder controllers, emergency stop and start switches.

[0020] This invention also provides a method for detecting the end geometry of hot-rolled strip steel based on image recognition. The method is implemented using the aforementioned system and includes:

[0021] A CMOS linear array sensor is used to perform nonlinear, staged measurements on the board material.

[0022] The parameters of the acquired images are calibrated using the Dréréni dynamic calibration CMOS linear array sensor and swarm intelligence optimization algorithm.

[0023] The acquired images are processed using image recognition algorithms, and the contour shape of the head and tail of the high-temperature steel plate is identified using image stitching technology.

[0024] Preferably, the method for performing nonlinear, staged measurements on the board material using a CMOS linear array sensor includes:

[0025] 3-1: Fix the calibration plate, install the checkerboard calibration plate on the translation platform, make the plane of the checkerboard calibration plate perpendicular to the camera optical axis, and perform multiple calibrations using a laser level;

[0026] 3-2: Turn on the CMOS linear array sensor with a filter on the main crossbeam, and at the same time turn on the laser emitter and the bar light source, so that the laser emitter emits uniform rays, and the CMOS linear array sensor collects the head and tail contour images of the conveyed plate on the roller conveyor when the bar light source is turned on.

[0027] 3-3: Use a CMOS linear array sensor to acquire static calibration board images. By continuously changing the position and orientation of the chessboard grid, take 20-25 related chessboard grid images.

[0028] 3-4: Using the laser emission point as the corner point and the linear emission line as the dividing line, divide the plate into regions A, B, C, and D along the direction perpendicular to the front and rear short crossbeams;

[0029] 3-5: Move the checkerboard calibration board to an initial position within the camera's field of view, extract the corner coordinates on the static checkerboard calibration board, and photograph the corner coordinates of each segmented area, recording them as the reference point (x0, y0);

[0030] 3-6: Start the rollers and control them with the industrial computer and speed controller to make the roller conveyor move at a constant speed.

[0031] 3-7: Collect the corner coordinates of the checkerboard calibration plate under dynamic conditions, and extract the corresponding dynamic coordinates along the direction of the roller movement according to the statically divided detection area, and record them as (x′, y′);

[0032] 3-8: Repeat steps 3-1 to 3-7 to perform multi-location and multi-frequency data acquisition;

[0033] 3-9: Calculate the offset Δx and offset angle θ in the image. The offset Δx is obtained by Δx = x′ - x0, and the offset angle θ is obtained by Δx = y·tanθ·scaling factor through the distortion model, where y represents the direction of the vertical coordinate in the image or coordinate system.

[0034] Preferably, the method for parameter calibration of acquired images using the Dréréni dynamic calibration of a CMOS linear array sensor and swarm intelligence optimization algorithm includes:

[0035] 3-10-1: Assume that feature point P is given a time dimension t and a motion parameter v based on the area array camera imaging model;

[0036] 3-10-2: An object moves at velocity V along the world coordinate system Y w The object point's coordinates are updated to Y′ at time t during axis motion. w =Y W +V·t is mapped to the camera coordinate system through the rotation matrix R and the translation vector T:

[0037]

[0038] Among them, X c Y c Z c Let Z be the camera coordinate system, with the origin at the optical center. C The axis points to the imaging plane, X′ w 、Y′ w Z′ w To update the X, Y, and Z axes of the world coordinate system;

[0039] 3-10-3: P(X) in camera coordinates C ,Y C Z C The mapping relationship from the coordinates of a point to the image point p(x, y) in the image coordinate system is derived using similar triangles:

[0040]

[0041] Where f is the focal length of the camera;

[0042] In the imaging model of the steel plate linear array camera, only the projected coordinates in the x-direction are acquired, and the longitudinal y-direction corresponds to the integral over time t:

[0043]

[0044] 3-10-4: Assuming the origin o of the image coordinate system is located in the pixel coordinate system (u0, u0), and ignoring the non-rectangular influence of the image sensor pixel array, the image coordinates (x, y) and pixel coordinates (u, v) have the following relationship:

[0045]

[0046] 3-10-5: From image coordinate system to pixel coordinate system and the detected object along Y... w The direction is moving at velocity V, and the total scanning time is T. total Then the vertical pixel coordinate v is:

[0047]

[0048] Y′ w =Y W Substitute +V·t into the camera coordinate system, and use t=v / f line Eliminating the time variable, the imaging model of the linear scan camera is derived:

[0049]

[0050] Where u is the horizontal pixel coordinate of the image; v is the vertical pixel coordinate of the image; f x f is the focal length of the camera in the x-direction, i.e., the horizontal direction; line For line frequency; r 11 r 12 r 13 r 31 r 32 r 33 V is the rotation matrix element; V is the camera's velocity in the Y direction; t x The x-component of the translation vector; t z The z-component of the translation vector.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] This invention employs a CMOS linear array sensor to perform nonlinear, staged measurements on the sheet metal. It uses Draréni dynamic calibration of the CMOS linear array sensor and a swarm intelligence optimization algorithm for parameter calibration. Image stitching technology is used to identify the contour shape of the head and tail of the high-temperature steel plate. Through these methods, the contour of the sheet metal can be obtained more accurately. Attached Figure Description

[0053] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a diagram showing the markings of steel plate locations in an embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram illustrating the principle in an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram of a hot-rolled strip end geometry detection system based on image recognition, as described in an embodiment of the present invention.

[0057] Figure 4 This is a schematic diagram of the dynamic imaging model in an embodiment of the present invention;

[0058] Figure 5 This is a schematic diagram illustrating the dynamic calibration principle in an embodiment of the present invention;

[0059] In the diagram: 1. Secondary crossbeam; 2. Main crossbeam; 3. First CMOS linear array sensor; 8. Second...

[0060] CMOS linear array sensor; 4. First strip light source; 9. Second strip light source; 5. Roller; 6. Anti-slip pad; 7. Short crossbeam; 10. High temperature steel plate; 11. Support base; 12. Sprocket. Detailed Implementation

[0061] 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.

[0062] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0063] Example 1

[0064] like Figure 3As shown, this invention provides an image recognition-based hot-rolled strip end geometry detection system, mainly comprising a fixed support module, a PLC control system, an image sensor acquisition module, a roller conveyor speed control system, a cooling system, and a data acquisition and classification interface. The fixed support module is a crucial component ensuring stable image acquisition by the camera and light source during roller conveyor transport. It mainly includes eight vertical camera supports, four short crossbeams (7), one 2-meter main crossbeam (2), two 1.2-meter secondary crossbeams (1), eight anti-slip pads (6), and 28 fixing angle irons. The vertical camera supports are connected by the short crossbeams (7), and fixed at the joints by angle irons. The short crossbeams (7) are connected to the secondary crossbeams (1) by angle irons, and the secondary crossbeams (1) are connected to the main crossbeams (2) by angle irons and bolts. The cameras are mounted on the main crossbeams (2) using camera mounting blocks and shock-absorbing pads. The vertical camera supports are in direct contact with the ground via the anti-slip pads (6), preventing high-speed contact. The high-temperature steel plate 10 experiences image distortion due to vibration during roller conveying. The PLC control system mainly includes a host computer, communication line, acquisition card, roller speed feedback module, camera acquisition command transmission module, and image information data transmission module. It is mainly responsible for the power supply and control of the entire detection system and the rapid transmission of feedback signals. The image sensing acquisition module is the main component for acquiring images of the calibration plate and the high-temperature steel plate. The quality of the acquired images is an important indicator affecting data processing. It mainly includes a first CMOS linear array sensor 3, a second CMOS linear array sensor 8, a first laser emitter, a second laser emitter, and a first strip laser sensor. Light source 4, second strip light source 9, camera mounting block, anti-slip pads 6, three-axis industrial camera gimbal, camera image stabilization stabilizer, first CMOS linear array sensor 3, and second CMOS linear array sensor 8 are mounted on the three-axis industrial camera gimbal. The three-axis industrial camera gimbal is mounted on the 2-meter main beam 2 via the camera mounting block. The camera is equipped with an image stabilization stabilizer to ensure stable image acquisition and prevent image distortion. The first strip light source 4 and second strip light source 9 are mounted on the support base 11 via light source clamps and support frames. A laser level measuring instrument is used to ensure that the first CMOS linear array sensor 3 and the second CMOS linear array sensor 8 are on the same horizontal plane. The irradiation area is positioned within the target detection area by adjusting the offset angle of the light source fixture; the first and second laser emitters are mounted on the 2-meter main beam 2 via camera fixing blocks to ensure that the camera detects the edge contour dimensions of the target object; the data acquisition and classification interface is used to classify the acquired data; the cooling system mainly includes water-cooled pipes and compressed air curtains. The cooling water unit is connected to the water-cooled plate via flexible hoses and to the plate heat exchanger via rigid pipes. The entire cooling water unit is powered by a circulating pump. The compressed air curtain is mainly connected via nozzles and air supply hoses. A honeycomb rectifier is installed at the nozzle outlet to cool the detector and prevent damage to the instrument.The roller conveyor speed control system mainly includes rollers 5, sprockets 12, chains, transmission chains, a 200W servo motor, an intelligent digital display speed controller, diffuse reflection photoelectric switches, encoders, encoder controllers, and emergency stop and start switches. The sprockets and chains include driving sprockets and chains, and driven sprockets and chains. The driving sprockets and chains are connected to the motor, and the driven sprockets and chains are connected to the driving sprockets and chains via chains. Rollers 5 are connected via driven sprockets and chains. The 200W servo motor, intelligent digital display speed controller, diffuse reflection photoelectric switches, encoders, encoder controllers, and emergency stop and start switches are all mounted on the roller conveyor support. The roller conveyor speed control system quickly feeds back the detected target speed and the actual running speed of the roller conveyor to the PLC control system through the diffuse reflection photoelectric switches and encoders, reducing the signal delay of the camera acquiring images.

[0065] Example 2

[0066] This embodiment describes a method for detecting the end geometry of hot-rolled strip steel based on image recognition. The method is implemented using the system described in Embodiment 1. Specifically, the head and tail shape detection of the steel plate on the roller conveyor is achieved through the following steps:

[0067] 1. Establishment of the testing platform;

[0068] 1-1: Connect roller 5 with bolts, and install the drive sprocket and chain on the servo motor to test the stability of the motor drive;

[0069] 1-2: Install rollers 5 at 50mm intervals, install driven sprockets and chains on the drive side of rollers 5, start the motor, and check that rollers 5 are running normally;

[0070] 1-3: A speed controller is installed on the outer layer of roller 5 to ensure that the motor operates within the normal speed range. At the same time, anti-slip pads 6 are installed on the support chassis of roller 5 to ensure the stability of the steel plate as it rotates on roller 5.

[0071] 1-4: Four sets of vertical camera brackets are connected by bolts. Each set of vertical camera brackets is connected by a short crossbeam 7. The short crossbeam 7 and the vertical camera brackets are fixed by angle iron to ensure the stability of each set of vertical camera brackets.

[0072] 1-5: Adjust the height of the vertical camera brackets, measure the installation height of each set of vertical camera brackets, and make the height of each set of vertical camera brackets consistent to ensure that the cameras are installed on the same horizontal plane later.

[0073] 1-6: A 1.2-meter-long secondary crossbeam 1 connects the two sets of vertical camera supports. The crossbeam and the vertical camera supports are fixed with angle iron. The secondary crossbeam 1 is measured with a level to ensure that the two secondary crossbeams 1 are on the same horizontal plane. At the same time, anti-slip pads 6 are installed at the bottom of the four sets of vertical camera supports to ensure the stability of the vertical camera supports.

[0074] 1-7: Use a laser level measuring instrument to recalibrate the four sets of vertical camera supports, the two sets of short crossbeams 7 and the one set of secondary crossbeams 1 to ensure that the vertical camera supports are on the same horizontal plane;

[0075] 1-8: Install the linear array CMOS sensor on the main crossbeam 2 using the camera mounting block to ensure that the linear array CMOS sensor is on the same horizontal plane; install the image stabilizer and shock-absorbing pad between the camera mounting block and the linear array CMOS sensor to ensure the stability of the image captured by the camera.

[0076] 1-9: Support bases 11 are installed on both sides of roller 5. By adjusting the height of the support frame and the angle of the light source fixture, it is ensured that the light source illuminates the entire steel plate being inspected. A 200*200mm anti-slip base is installed at the bottom of the light source support rod to ensure the stability of the light source.

[0077] 1-10: Using the movement trajectory of the steel plate as a reference direction, the camera mounting bracket is divided into a front bracket and a rear bracket.

[0078] 2. Debugging of each module;

[0079] 2-1: Use a laser level to recalibrate whether the vertical camera bracket is on the same horizontal plane. If the bracket is not on the same horizontal plane, adjust the height of the bracket to ensure that the relative position of the vertical camera bracket is stable. Use the laser level to measure whether the CMOS linear array sensor and the bar light source are on the same horizontal plane. The distance between the bar light source and the CMOS linear array sensor is 20cm. At the same time, adjust the rotation angle of the bar light source so that the illumination area of ​​the bar light source is at 35° with the optical axis of the linear array camera to ensure that the light source illuminates the detection target area.

[0080] 2-2: Adjust the position and angle of the laser emitter. According to the size of the steel plate, fix the laser on the main crossbeam 2 at a distance of 450mm from the camera. The angle between the laser emission angle and the optical axis of the camera is 30°. The coverage margin of the laser spot extends 15% outward from the detection area. Use a low-power laser and support the laser spot on the cold steel plate to ensure that the light emitted by the laser emitter is evenly irradiated on the high-temperature steel plate 10 being detected.

[0081] 2-3: Connect the power supply to the motor in roller 5, connect the power cord to the intelligent digital display speed controller, set the motor rotation direction to the direction of steel plate movement and set it to forward rotation. In the menu settings, set the forward rotation speed to 0.5m / s, the forward acceleration start time to 0.1s, the forward stop and slow deceleration time to 0.1s, and the alarm function to be enabled. The speed adjustment method is panel adjustment. Start the speed controller and motor, and check whether the parameters are set successfully.

[0082] 2-4: Connect the encoder to the industrial computer via the encoder controller, and connect the diffuse reflection photoelectric switch signal line at the same time. Turn on the industrial computer and rotate the roller conveyor in the forward direction. Test whether the encoder and photoelectric switch signals quickly reflect the rotation speed of the roller conveyor in the Arduino IDE interface. If a speed value appears, the encoder and photoelectric switch debugging is complete.

[0083] 2-5: Install the image acquisition card on the industrial computer to ensure that the acquired images and data can be processed quickly;

[0084] 2-6: Install the lens, connect the camera power cable and I / O cable;

[0085] 2-7: Install the camera acquisition software on the industrial control computer, open the software, and manually adjust the lens aperture and focal length, with the lens focal length set to 20mm. Simultaneously, check the image clarity in the MVS-STD software to ensure the acquired image of the high-temperature steel plate 10 is not distorted and to guarantee the clarity of the acquired image.

[0086] 2-8: Turn on the industrial control computer and check again in the camera acquisition software interface whether the encoder, photoelectric switch, frequency converter and CMOS linear array sensor are properly connected; check whether each module controlled by the industrial control computer is running normally.

[0087] like Figure 1 As shown, 3. Image acquisition and parameter calibration;

[0088] 3-1: Fix the calibration plate, install the checkerboard calibration plate on the translation platform, ensure that its plane is perpendicular to the camera optical axis, and perform multiple calibrations using a laser level;

[0089] 3-2: Turn on the CMOS linear array sensor with a filter on the 2-meter main beam 2, and turn on the laser emitter and the bar light source at the same time, so that the laser emitter emits uniform rays, and the CMOS linear array sensor collects the head and tail contour images of the conveyed plate on the roller conveyor when the bar light source is turned on.

[0090] 3-3: Use a CMOS linear array sensor to acquire static calibration board images. The checkerboard calibration board is completely imaged in the camera's field of view. The calibration board is centered and parallel to the imaging plane. The image is taken by rotating ±30° within the plane. The checkerboard calibration board is taken with the camera's optical axis as the reference and the pitch angle is 45°. The corner points of the checkerboard in the captured image are clear. By continuously adjusting the position and angle of the checkerboard, 20-25 relevant checkerboard images are acquired.

[0091] 3-4: Using the laser emission point as the corner point and the linear emission line as the dividing line, divide the plate into regions A, B, C, and D along the direction perpendicular to the front and rear short crossbeams 7;

[0092] 3-5: Remove the damaged images with unclear corner points from the acquired chessboard images, retain the high-resolution and undamaged chessboard calibration board images, and use the Camera Calibrator in the MATLAB software toolbox to extract the corner coordinates on the static chessboard calibration board. Extract the corner coordinates of each segmented region and record them as the reference point (x0, y0).

[0093] 3-6: Start the motor switch of roller 5, adjust the motor speed through the frequency converter, turn the rotation button to 40, and stabilize the speed at 0.5m / s. Install a photoelectric switch at the bottom of the roller conveyor. The photoelectric switch displays the actual speed of the roller conveyor through the Arduino IDE interface to ensure that the roller conveyor moves at a uniform speed.

[0094] 3-7: Collect the corner coordinates of the checkerboard calibration plate under dynamic conditions, and extract the corresponding dynamic coordinates along the direction of the roller movement according to the statically divided detection area, and record them as (x′, y′);

[0095] 3-8: Repeat steps 3-1 to 3-7 to perform multi-location and multi-frequency data acquisition (e.g. Figure 5 (as shown);

[0096] 3-9: Calculate the offset and offset angle θ in the image. Obtain the offset according to Δx=x′-x0, and obtain the offset angle θ through the distortion model Δx=y·tanθ·scaling factor;

[0097] 3-10: Perform parameter calibration on the acquired images (results of nonlinear phased measurement) to ensure the accuracy of the obtained steel plate visual recognition;

[0098] Linear scan camera parameter calibration is used to accurately establish the mapping relationship between one-dimensional image coordinates and two-dimensional world coordinates along the motion direction. Therefore, a linear scan camera imaging model is established based on the area scan camera imaging model:

[0099] 3-10-1: Assume that feature point P is augmented with a time dimension t and a motion parameter v in the area scan camera imaging model. The area scan camera imaging model is expressed as:

[0100]

[0101] Among them, X c Y c Z c Let Z be the camera coordinate system, with the origin at the optical center. C The axis points to the imaging plane.

[0102] 3-10-2: An object moves at velocity V along the world coordinate system Y w The object point's coordinates are updated to Y′ at time t during axis motion. w =Y W +V·t is mapped to the camera coordinate system through the rotation matrix R and the translation vector T:

[0103]

[0104] Where, X′ w 、Y′ w Z′ w This refers to the X, Y, and Z axes of the updated world coordinate system.

[0105] 3-10-3: P(X) in camera coordinates C ,Y C Z C The mapping relationship from the coordinates of a point to the image point p(x, y) in the image coordinate system can be derived using similar triangles:

[0106]

[0107] Where f is the focal length of the camera;

[0108] In the imaging model of the steel plate linear array camera, only the projected coordinates in the x-direction are acquired, and the longitudinal y-direction corresponds to the integral over time t:

[0109]

[0110] 3-10-4: Assuming the origin o of the image coordinate system is located in the pixel coordinate system (u0, v0), and ignoring the non-rectangular influence of the image sensor pixel array, the image coordinates (x, y) and pixel coordinates (u, v) have the following relationship:

[0111]

[0112] 3-10-5: From image coordinate system to pixel coordinate system (CMOS linear array sensor only retains one-dimensional projection in the u direction) and the detected object along the Y direction. w The direction is moving at velocity V, and the total scanning time is T. total Then the vertical pixel coordinate v is:

[0113]

[0114] Y′ w =Y W Substitute +V·t into the camera coordinate system, and use t=v / f line Eliminating the time variable, the imaging model of the linear scan camera is derived:

[0115]

[0116] Where u is the horizontal pixel coordinate of the image (in pixels); v is the vertical pixel coordinate of the image (in pixels); f x f is the focal length of the camera in the x-direction (horizontal direction); line Line frequency (in Hz); r 11 r 12 r 13 r 31 r 32 r 33 V represents the rotation matrix element; V is the camera's velocity in the Y direction (in m / s); t x The x-component of the translation vector (unit: m); t z The z-component of the translation vector.

[0117] 3-11: Finally, the imaging model of the above-mentioned linear scan camera (such as...) Figure 4 (As shown) The head and tail shape contours of the high-temperature steel plate 10 are accurately detected using an intelligent optimization algorithm. The specific implementation process is as follows:

[0118] 3-11-1: The displacement of the steel plate is measured in real time by encoding. When the steel plate moves to a preset small distance, the encoder sends a pulse signal to trigger the camera to expose once. The CMOS linear array sensor acquires a line image of the current cross section of the steel plate each time it is triggered. Through the continuously acquired line images, a complete two-dimensional image of the steel plate is constructed in the host computer.

[0119] 3-11-2: When positioning the head and tail areas of the steel plate, when the starting point of the steel plate head appears, the CMOS linear array sensor continuously collects data to completely cover the head area, and continues scanning until the starting point of the tail appears, and then continuously collects data to completely cover the tail area.

[0120] 3-11-3: From the stitched image, the head and tail regions are extracted. The mature Gaussian filter is used to remove noise. At the same time, the existing Canny operator is applied to highlight the boundary between the steel plate and the background. The extracted regions are adaptively binarized, and the existing contour tracking algorithm is used to extract the contours representing the head and tail.

[0121] 3-11-4: Define the fitness function through the imaging model of the linear array camera, and use the improved WOA-LM optimization algorithm to construct parameterized curves and fit the measured contour point cloud to obtain the overall contour including the head and tail of the steel plate.

[0122] The above steps completed the scanning of the board's outline.

[0123] like Figure 2 As shown, to further complete the contour detection of the 10 ends of the high-temperature steel plate, the principle of this invention is mainly introduced as follows:

[0124] This invention employs a segmented acquisition and phased optimization method for scanning. During the parameter calibration stage, the dynamic imaging principle of a CMOS linear array sensor is introduced, deriving the imaging model of the linear array camera. By incorporating motion information, a combination of Draréni dynamic calibration of the CMOS linear array sensor and a swarm intelligence optimization algorithm is used to obtain more accurate calibration parameters, thereby improving calibration accuracy and efficiency. Finally, image stitching and edge feature extraction are used to detect the head and tail contours of the steel plate. During the steel plate's transport, the sensor identifies four detection areas: A, B, C, and D. When the steel plate head enters areas A and B via a photoelectric switch and laser emitter, the second CMOS linear array sensor 8 detects the shape of the steel plate head through signal feedback. At this time, the relative motion direction of the camera is strictly perpendicular to the pixel array direction. The reference point that first enters the camera's detection area within areas A and B is denoted as P1, and the reference point detected by the camera in the irregular recessed area of ​​the plate head is denoted as P3. The mapping point of P3 perpendicular to the steel plate's motion direction is denoted as P′3. At this point, the straight-line distance P1 perpendicular to the pixel array is the detection distance of the plate head not entering the irregular area, denoted as l1; it can be obtained by the following formula: l1=H·tanα;

[0125] When the second CMOS linear array sensor 8 acquires images of regions A and B, and the contour dimensions on both sides of the steel plate head tend to be uniform, a theoretical straight line y is drawn on both sides along the direction perpendicular to the pixel array. The corner point of y and curve P1P3 is P2, and the mapping point of P2 along the direction perpendicular to the movement of the steel plate is recorded as P′2. P′2 is the critical point for evaluating irregular regions. When the linear array camera performs detection on the plate head, when the camera imaging area scans the irregular region P′3, the detection distance of the camera in the irregular region of the plate head is recorded as S1. The formula is: S1=H·tan(α+β+γ), where H is the height of the camera from the reference plane, and α, β, and γ are the angles between the camera optical axis and the mapping point.

[0126] Furthermore, the length of the irregular region actually detected by the second CMOS linear array sensor 8 at the head of the plate is: l2 = S1 - l1; at the same time, when the two local areas C and D of the plate move to the rear CMOS linear array sensor, the length of the irregular region at the tail of the plate can be obtained by repeating the above principle to scan the irregular region at the tail of the plate.

[0127] By cooperating with the first CMOS linear array sensor 3 and the second CMOS linear array sensor 8, the surface contour of the head and tail of the board is scanned, and the contour curve of the board is accurately depicted by image stitching.

[0128] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for detecting the end geometry of hot-rolled strip steel based on image recognition, characterized in that, The method utilizes an image recognition-based hot-rolled strip end geometry detection system, which includes: a fixed support module, a PLC control system, an image sensor acquisition module, a roller speed control system, a cooling system, and a data acquisition and classification interface. The fixed support module is used to support the image sensing and acquisition module; The image sensing and acquisition module is used to acquire images of the calibration plate and the high-temperature steel plate; The data collection and classification interface is used to classify the collected data; The PLC control system is used for power supply and control of the entire detection system, and is responsible for signal transmission of the feedback system. The cooling system is used to cool the detector. The roller conveyor speed control system is used to feed back the detected target speed and the actual running speed of the roller conveyor to the PLC control system. The method includes: A CMOS linear array sensor is used to perform nonlinear, staged measurements on the board material. The parameters of the acquired images are calibrated using the Dréréni dynamic calibration CMOS linear array sensor and swarm intelligence optimization algorithm. The acquired images are processed using image recognition algorithms, and the contour shape of the head and tail of the high-temperature steel plate is identified using image stitching technology. Methods for nonlinear, staged measurement of sheet metal using CMOS linear array sensors include: 3-1: Fix the calibration plate, install the checkerboard calibration plate on the translation platform, make the plane of the checkerboard calibration plate perpendicular to the camera optical axis, and perform multiple calibrations using a laser level; 3-2: Turn on the CMOS linear array sensor with a filter on the main crossbeam, and at the same time turn on the laser emitter and the bar light source, so that the laser emitter emits uniform rays, and the CMOS linear array sensor collects the head and tail contour images of the conveyed plate on the roller conveyor when the bar light source is turned on. 3-3: Use a CMOS linear array sensor to acquire static calibration board images. By continuously changing the position and orientation of the chessboard grid, take 20-25 related chessboard grid images. 3-4: Using the laser emission point as the corner point and the linear emission line as the dividing line, divide the plate into regions A, B, C, and D along the direction perpendicular to the front and rear short crossbeams; 3-5: Move the checkerboard calibration board to an initial position within the camera's field of view, extract the corner coordinates of the static checkerboard calibration board, and photograph the corner coordinates of each segmented region, recording them as reference points. ; 3-6: Start the rollers and control them with the industrial computer and speed controller to make the roller conveyor move at a constant speed. 3-7: Collect the corner coordinates of the checkerboard calibration plate under dynamic conditions. Based on the statically divided detection area, extract the corresponding dynamic coordinates along the direction of the roller conveyor movement, and record them as follows: ; 3-8: Repeat steps 3-1 to 3-7 to perform multi-location and multi-frequency data acquisition; 3-9: Calculate the offset in the image and offset angle .

2. The method according to claim 1, characterized in that, The fixed support module includes: 8 vertical camera brackets, 4 short crossbeams, 1 2-meter main crossbeam, 2 1.2-meter secondary crossbeams, 8 anti-slip pads, and 28 fixed angle irons; The vertical camera brackets are connected by short crossbeams, which are fixed with angle irons. The short crossbeams are connected to the secondary crossbeams by angle irons, and the secondary crossbeams are connected to the main crossbeams by angle irons and bolts. The cameras are mounted on the main crossbeams by camera mounting blocks and shock-absorbing rubber pads. The vertical camera brackets are in direct contact with the ground by anti-slip feet.

3. The method according to claim 2, characterized in that, The PLC control system includes: a host computer, a communication line, a data acquisition card, a roller speed feedback module, a camera acquisition command transmission module, and an image information data transmission module.

4. The method according to claim 3, characterized in that, The image sensing and acquisition module includes: a first CMOS linear array sensor, a second CMOS linear array sensor, a first laser emitter, a second laser emitter, a first bar light source, a second bar light source, a camera mounting block, anti-slip pads, a three-axis industrial camera gimbal, and a camera image stabilization stabilizer. The first and second CMOS linear array sensors are mounted on a three-axis industrial camera gimbal. The three-axis industrial camera gimbal is mounted on the main beam via a camera mounting block. The first and second strip light sources are mounted on a support base via light source clamps and support frames. A laser level measuring instrument is used to ensure that all CMOS linear array sensors and all light sources are on the same horizontal plane. The offset angle of the light source clamps is adjusted to ensure that the illumination area is within the target detection area. The first and second laser emitters are mounted on the main beam via camera mounting blocks to ensure that the camera detects the edge contour dimensions of the target object.

5. The method according to claim 1, characterized in that, The cooling system includes: water-cooled pipes and compressed air curtain.

6. The method according to claim 1, characterized in that, The roller conveyor speed control system includes: rollers, sprockets and chains, transmission chains, 200W servo motors, intelligent digital display speed controllers, diffuse reflection photoelectric switches, encoders, encoder controllers, emergency stop and start switches.

7. The method according to claim 1, characterized in that, The method for parameter calibration of acquired images using Draréni dynamic calibration of a CMOS linear array sensor and swarm intelligence optimization algorithms includes: 3-10-1: Assuming Feature Points A time dimension was introduced into the imaging model of an area array camera. and motion parameters ; 3-10-2: An object moves at a velocity... Along the world coordinate system Axis motion, time The coordinates of the time point are updated to By rotation matrix Translation vector Mapped to camera coordinate system: ； in, , , Let the camera coordinate system be the origin at the optical center. The axis points to the imaging plane. , , To update the world coordinate system X, Y, Z axis; 3-10-3: Camera Coordinates Point coordinates to image point in image coordinate system The mapping relationship is derived through similar triangles: ； in, The focal length of the camera; In the imaging model of the steel plate linear array camera, only data is collected. Projected coordinates in the direction, longitudinal Corresponding time Integrals: ； 3-10-4: Assume the origin of the image coordinate system Located in pixel coordinate system Ignoring the non-rectangular influence of the image sensor pixel array, image coordinates With pixel coordinates The following relationship exists: ； 3-10-5: From image coordinate system to pixel coordinate system and the detected object along... Direction by speed Movement, total scan time is Then the vertical pixel coordinate for: ； ； Will Substitute the camera coordinate system and utilize The imaging model of the linear array camera is derived as follows: ； in, The horizontal pixel coordinates of the image; The vertical pixel coordinates of the image; For the camera x The direction refers to the focal length in the horizontal direction; For line frequency; , , , , , Elements of the rotation matrix; The camera's velocity in the Y direction; Translation vector x Quantity; Translation vector z Quantity, p To detect the actual physical distance of a target per unit pixel in the direction of motion.

Images