Tobacco shred crimpness measuring device and method based on three-dimensional platform

Abstract

The invention discloses a tobacco shred crimpness measuring device and method based on a three-dimensional platform, and the device comprises a conveyor belt conveying mechanism which is used for conveying a tobacco shred sample and moving the tobacco shred sample to a detection region of the conveyor belt conveying mechanism; the X-axis and Y-axis electric radian table is fixedly connected with the conveyor belt conveying mechanism and used for achieving angular deviation of the conveyor belt conveying mechanism in the X-axis direction and the Y-axis direction; the illumination system and the camera are mounted above the detection area, the illumination system is used for projecting the edge contour of the cut tobacco sample, the camera is used for obtaining cut tobacco sample images at different angles through the XY-axis electric radian table, and the illumination system and the camera are synchronously controlled to ensure that the image quality is consistent; the negative pressure adsorption system is used for ensuring that the cut tobacco samples are kept stable in the conveying and shooting processes; and the data processing unit is used for performing image processing on the tobacco shred sample images at different angles to obtain the crimpness of the tobacco shred sample, so that high precision, automation and repeatability of tobacco shred crimpness detection are realized.

Description

Technical Field

[0001] This invention relates to the fields of intelligent detection technology and tobacco machinery technology, specifically to a device and method for measuring the curvature of tobacco shreds based on a three-dimensional platform. Background Technology

[0002] The curl of tobacco shreds is an important indicator characterizing the appearance and processing quality of tobacco shreds, and it has a significant impact on the filling density, ventilation resistance, and combustion performance of cigarettes. Currently, the measurement of tobacco curl mainly includes three methods: manual inspection, two-dimensional image analysis, and contact mechanical measurement. 1. The manual inspection method relies on experienced personnel to observe and judge the degree of curling under a microscope, which has problems such as strong subjectivity, poor repeatability and low efficiency (only 10-15 samples per hour). 2. Two-dimensional image analysis method acquires planar projection images through a top-view camera, but it cannot identify overlapping tobacco shreds and ignores the spatial curvature characteristics of the tobacco shreds, resulting in measurement results that are 15%-30% lower than the true values; 3. The contact mechanical measurement method measures the deformation by applying tensile force to calculate the curvature, but mechanical contact can easily change the original shape of the tobacco, causing a measurement error of 8%-12%, and the detection efficiency is low.

[0003] Existing methods for detecting the curliness of tobacco shreds still have significant shortcomings in terms of accuracy, stability, and adaptability, mainly in the following aspects: 1. Limitations in measurement dimensionality. Existing two-dimensional imaging technology can only acquire planar projection features and cannot capture the three-dimensional spatial curvature and twist angle of tobacco shreds (resulting in the loss of more than 30% of feature information). 2. Measurement accuracy defects. Manual visual inspection is affected by subjective factors, and the measurement results of different operators can vary by 20-25%. Contact measurement is affected by sample deformation caused by mechanical stress, and the measured data deviates from the true value by 8-12%.

[0004] 3. Efficiency bottleneck. Traditional methods can only measure 1-3 tobacco shreds at a time (taking 30-60 seconds), which cannot meet the testing needs of modern production lines that can process 500+ tobacco shreds per minute.

[0005] 4. Lack of data value. Existing methods only output a single curl value and lack key data for process optimization such as spatial curvature distribution and anisotropy; Existing technologies generally suffer from insufficient accuracy (significant errors in thin cigarettes), efficiency bottlenecks (difficult to match the testing requirements of 300-500 grams per minute in modern production lines), and lack of data dimensions (lacking key parameters such as spatial curvature). Therefore, there is an urgent need for an automated testing device and method that can achieve multi-angle acquisition and spatial morphology reconstruction of tobacco shreds under non-contact conditions to improve measurement accuracy and efficiency. Summary of the Invention

[0006] To address the problems of insufficient measurement dimensions, low accuracy, and poor detection efficiency in existing technologies, this invention aims to provide a tobacco curling degree measurement device and method based on a three-dimensional platform. Through the collaborative work of multi-angle blue light imaging, XY-axis electric arc stage, and image fusion algorithm, it achieves accurate measurement of the spatial curling characteristics of tobacco.

[0007] The present invention adopts the following technical solution: A device for measuring the curvature of tobacco shreds based on a three-dimensional platform, comprising: Conveyor belt conveyor mechanism, lighting system, camera, negative pressure adsorption system, XY axis electric arc stage and data processing unit; The conveyor belt conveyor mechanism is used to transport tobacco samples and move them to the detection area of ​​the conveyor belt conveyor mechanism; The XY-axis electric arc stage is fixedly connected to the conveyor belt conveying mechanism and is used to realize the angular offset of the conveyor belt conveying mechanism along the X and Y axes. The lighting system and camera are mounted above the detection area. The lighting system is used to highlight the edge contour of the tobacco sample, and the camera is used to obtain images of the tobacco sample at different angles through the XY-axis electric arc stage. The lighting system and camera are controlled synchronously to ensure consistent image quality. The negative pressure adsorption system is used to ensure that the tobacco sample remains stable during transportation and imaging. The data processing unit is used to perform image processing on tobacco sample images from different angles to obtain the curvature of the tobacco sample.

[0008] Furthermore, the lighting system is specifically a blue light lighting system with a wavelength of 415 nanometers, which has good scattering suppression performance and can enhance the contrast of the tobacco edge contour; the camera is a binocular industrial camera, specifically a binocular industrial camera with a resolution of 2592×1944.

[0009] Furthermore, the conveyor belt of the conveyor belt conveying mechanism adopts a composite structure, the composite structure includes a base belt and a breathable layer, the base belt has a number of annular slits evenly distributed on it, the breathable layer covers the surface of the base belt, and the surface of the breathable layer has a number of micropores evenly distributed on it.

[0010] Furthermore, the baseband is made of antistatic black substrate, which effectively reduces optical reflection interference; the breathable layer is made of food-grade non-woven fabric, and the micropore diameter is 1-100μm.

[0011] Furthermore, the negative pressure adsorption system includes a vacuum pump, a gas path, and a gas pressure sensor; The air pressure sensor is used to detect the air pressure inside the conveyor belt in real time. When the air pressure is not at the set threshold, the vacuum pump automatically adjusts to maintain a constant adsorption force of the conveyor belt on the tobacco sample, ensuring the stability of the shooting. One end of the air path is connected to the inside of the conveyor belt conveying mechanism, and the other end is connected to the vacuum pump. The vacuum pump draws a vacuum through the air path, the annular slit, and the micropores evenly distributed on the breathable layer to form a controllable negative pressure field, so that the tobacco sample is always in close contact with the surface of the conveyor belt during the operation of the conveyor belt conveying mechanism and during shooting, preventing the attitude deviation caused by airflow disturbance or vibration. Preferably, the normal threshold range is −3 kPa to −6 kPa.

[0012] Furthermore, the camera is used to obtain images of tobacco samples at different angles via the XY-axis electric arc stage. Specifically, the XY-axis electric arc stage adjusts the tilt angle of the conveyor belt to generate five sets of view images: a top-down view, two sets of tilted view along the positive and negative X-axis directions respectively, and two sets of tilted view along the positive and negative Y-axis directions respectively.

[0013] An image processing method for a tobacco curl measurement device based on a three-dimensional platform, implemented using any of the aforementioned measurement devices, includes the following steps: S1. Coordinate calibration is performed on tobacco sample images from different angles to establish the mapping relationship between the world coordinate system and the camera coordinate system, providing a geometric basis for 3D reconstruction; S2. Perform stereo matching and feature calibration on tobacco sample images from different angles, and then fuse them based on the feature point matching algorithm to obtain point cloud data of tobacco sample images from different angles. Based on the point cloud data fusion, obtain a complete three-dimensional spatial point cloud model of tobacco. Preferably, post-process the complete three-dimensional spatial point cloud model of tobacco, including point cloud filtering and noise reduction, and thinning. S3. Extract the skeleton of the complete three-dimensional point cloud model of the tobacco shreds to obtain the centerline curve function; calculate the curvature of the tobacco shreds based on the centerline curve function.

[0014] Furthermore, in S1, the coordinate calibration specifically refers to: The mapping relationship between the world coordinate system and the camera coordinate system is established using the camera rotation matrix and translation vector; in, , , For camera coordinate system coordinates, , , Let R be the world coordinate system coordinates, and R be a 3×3 rotation matrix used to describe the attitude change (pitch / yaw / roll) of the camera coordinate system relative to the world coordinate system, including the camera's XYZ Euler angles. T be a 3×1 translation vector used to describe the position offset of the camera's optical center in the world coordinate system. The checkerboard calibration method is used to calculate (R,T) for any image, and to establish the mapping relationship between the camera coordinates and the world coordinates of any image.

[0015] Furthermore, in S2, the specific steps for performing stereo matching and feature calibration on tobacco sample images from different angles, and then fusing them based on a feature point matching algorithm to obtain point cloud data of tobacco sample images from different angles are as follows: The stereo matching is based on the depth value of the image from the i-viewpoint calculated using the parallax of a binocular industrial camera. The formula is: Where f is the focal length of the binocular industrial camera, B is the baseline distance of the binocular industrial camera, and d is the parallax of the binocular industrial camera. The features in the feature calibration specifically include the shape of the tobacco sample and the natural texture of the tobacco sample surface, wherein the natural texture includes fiber orientation and wrinkles; The image with calibrated features at different angles, the corresponding depth values, the mapping relationship between the corresponding camera coordinates and world coordinates are input into the feature point matching algorithm to obtain point cloud data of tobacco sample images at different angles; The point cloud data of the tobacco sample images from different angles were registered using the ICP algorithm: The ICP algorithm uses an error function to minimize the distance of the same feature point under different viewpoints, thus obtaining... , Optimal values ​​improve the registration accuracy of the ICP algorithm: Where n represents all n corresponding feature points. Let be the coordinates of the i-th feature point at any tilted viewpoint in the camera coordinate system. Let be the coordinates of the i-th feature point from the downward viewpoint in the camera coordinate system.

[0016] Furthermore, in S3, the skeleton of the complete three-dimensional point cloud model of the tobacco shreds is extracted to obtain the centerline curve function; the curvature of the tobacco shreds is calculated based on the centerline curve function, specifically as follows: The centerline curve function r(s) = [x(s), y(s), z(s)] of the complete 3D tobacco point cloud model is generated using a morphological refinement algorithm, where s is the arc length parameter. The arc length L of the tobacco shreds is obtained through the centerline curve function. Based on the centerline curve function r(s), the local curvature is calculated. : ; Based on local curvature Obtaining mean curvature : ; The average curvature of the output reflects the overall curl of the tobacco.

[0017] Beneficial effects: 1. The problem of missing curvature in three-dimensional space. A breakthrough was achieved in reconstructing three-dimensional space from two-dimensional projection, accurately extracting the spatial curvature parameters of tobacco shreds; 2. Contact interference problem. Non-contact multispectral structured light scanning is used to eliminate deformation errors caused by contact measurements; 3. Overlapping interference problem. Traditional algorithms cannot distinguish the outline of individual tobacco strands when tobacco strands are stacked. By introducing three-dimensional measurement angle parameters and an adaptive threshold model based on point cloud density, a three-dimensional curl index is established to reduce the error of the two-dimensional image method and improve the accuracy of curl measurement.

[0018] This invention achieves high precision, automation, and repeatability in the detection of tobacco curliness, providing reliable technical support for tobacco quality control. Attached Figure Description

[0019] Figure 1 This is a structural block diagram of a tobacco curl measurement device based on a three-dimensional platform according to an embodiment of the present invention; Figure 2 This is a three-dimensional structural diagram of the conveyor belt conveying mechanism, the XY-axis electric arc stage, and the negative pressure adsorption system of a tobacco curl measurement device based on a three-dimensional platform according to an embodiment of the present invention. Figure 3 This is a diagram of a tobacco curl measurement device based on a three-dimensional platform according to an embodiment of the present invention after a 20-degree offset. Figure 4 This is a cross-sectional view of a conveyor belt according to an embodiment of the present invention; Figure 5 This is a perspective view of a conveyor belt according to an embodiment of the present invention; Figure 6 This is a flowchart of a method for measuring the curvature of tobacco shreds based on a three-dimensional platform according to an embodiment of the present invention; In the diagram: 1. Conveyor belt conveyor mechanism; 2. Negative pressure adsorption system; 3. XY axis electric arc stage; 4. Breathable layer; 5. Base belt. Detailed Implementation

[0020] Example 1: A tobacco curl measurement device based on a three-dimensional platform includes: a conveyor belt conveying mechanism 1, an illumination system, a camera, a negative pressure adsorption system 2, an XY axis electric arc stage 3, and a data processing unit; The conveyor belt conveyor mechanism 1 is used to transport tobacco samples and move the tobacco samples to the detection area of ​​the conveyor belt conveyor mechanism 1. The XY-axis electric arc stage 3 is fixedly connected to the conveyor belt conveying mechanism and is used to realize the angular offset of the conveyor belt conveying mechanism along the X and Y axes. The lighting system and camera are mounted above the detection area. The lighting system is used to highlight the edge contour of the tobacco sample, and the camera is used to obtain images of the tobacco sample at different angles through the XY-axis electric arc stage. The lighting system and camera are controlled synchronously to ensure consistent image quality. The negative pressure adsorption system 2 is used to ensure that the tobacco sample remains stable during transportation and imaging. The data processing unit is used to perform image processing on tobacco sample images from different angles to obtain the curvature of the tobacco sample.

[0021] Furthermore, the lighting system is specifically a blue light lighting system with a wavelength of 415 nanometers, which has good scattering suppression performance and can enhance the contrast of the tobacco edge contour; the camera is a binocular industrial camera, specifically a binocular industrial camera with a resolution of 2592×1944.

[0022] like Figure 4 and 5 As shown, the conveyor belt of the conveyor belt conveying mechanism 1 adopts a composite structure. The composite structure includes a base belt and a breathable layer. The base belt has 20 evenly spaced annular slits, each 0.5 mm wide and 25 mm apart. The breathable layer covers the surface of the base belt, completely covering the base belt below, forming a "breathable but not fiber-permeable" buffer layer to prevent tobacco from being sucked into the vacuum tube. The surface of the breathable layer has several micropores evenly distributed. The base belt is an anti-static black substrate, which effectively reduces optical reflection interference. The breathable layer is a food-grade non-woven fabric, and the micropore diameter is 100 μm.

[0023] The negative pressure adsorption system 2 includes a vacuum pump, an air path, and a pressure sensor. The pressure sensor is used to detect the air pressure inside the conveyor belt in real time. When the air pressure is not -4 kPa, the vacuum pump automatically adjusts to maintain a constant adsorption force of the conveyor belt on the tobacco sample, ensuring the stability of the imaging. The vacuum pump is an oil-free dry vacuum pump to avoid oil contamination of the tobacco. It is equipped with a vacuum tank for buffering, stabilizing the negative pressure, and preventing the tobacco from shifting due to suction fluctuations. One end of the air path is connected to the inside of the conveyor belt conveyor mechanism, and the other end is connected to the vacuum pump. The vacuum pump creates a controllable negative pressure field by drawing a vacuum through the air path, the annular slit, and the micropores evenly distributed on the breathable layer. This ensures that the tobacco sample remains in close contact with the surface of the conveyor belt during the operation of the conveyor belt conveyor mechanism 1 and during imaging, preventing attitude shift caused by airflow disturbance or vibration.

[0024] The camera is used to obtain images of tobacco samples at different angles via the XY-axis electric arc stage 3. Specifically, the XY-axis electric arc stage 3 adjusts the tilt angle of the conveyor belt to generate 5 sets of view images: a top view, two sets of tilted view along the positive and negative X-axis, and two sets of tilted view along the positive and negative Y-axis.

[0025] Example 2: An image processing method for a tobacco curl measurement device based on a three-dimensional platform, implemented using the measurement device described in Example 1, includes the following steps: S1. Coordinate calibration is performed on tobacco sample images from different angles to establish the mapping relationship between the world coordinate system and the camera coordinate system, providing a geometric basis for 3D reconstruction; S2. Perform stereo matching and feature calibration on tobacco sample images from different angles, and then fuse them based on the feature point matching algorithm to obtain point cloud data of tobacco sample images from different angles. Based on the point cloud data, fuse them to obtain a complete three-dimensional spatial point cloud model of tobacco. S3. Extract the skeleton of the complete three-dimensional point cloud model of the tobacco shreds to obtain the centerline curve function; calculate the curvature of the tobacco shreds based on the centerline curve function.

[0026] Furthermore, in S1, the coordinate calibration specifically refers to: The mapping relationship between the world coordinate system and the camera coordinate system is established using the camera rotation matrix and translation vector; in, , , For camera coordinate system coordinates, , , Let R be the world coordinate system coordinates, and R be a 3×3 rotation matrix used to describe the attitude change (pitch / yaw / roll) of the camera coordinate system relative to the world coordinate system, including the camera's XYZ Euler angles. T be a 3×1 translation vector used to describe the position offset of the camera's optical center in the world coordinate system. The checkerboard calibration method is used to calculate (R,T) for any image, and to establish the mapping relationship between the camera coordinates and the world coordinates of any image.

[0027] Furthermore, in S2, the specific steps for performing stereo matching and feature calibration on tobacco sample images from different angles, and then fusing them based on a feature point matching algorithm to obtain point cloud data of tobacco sample images from different angles are as follows: The stereo matching is based on the depth value of the image from the i-viewpoint calculated using the parallax of a binocular industrial camera. The formula is: Where f is the focal length of the binocular industrial camera, B is the baseline distance of the binocular industrial camera, and d is the parallax of the binocular industrial camera. The features in the feature calibration specifically include the shape of the tobacco sample and the natural texture of the tobacco sample surface, wherein the natural texture includes fiber orientation and wrinkles; The image with calibrated features at different angles, the corresponding depth values, the mapping relationship between the corresponding camera coordinates and world coordinates are input into the feature point matching algorithm to obtain point cloud data of tobacco sample images at different angles; The point cloud data of the tobacco sample images from different angles were registered using the ICP algorithm: The ICP algorithm uses an error function to minimize the distance of the same feature point under different viewpoints, thus obtaining... , Optimal values ​​improve the registration accuracy of the ICP algorithm: Where n represents all n corresponding feature points. Let be the coordinates of the i-th feature point at any tilted viewpoint in the camera coordinate system. Let be the coordinates of the i-th feature point from the downward viewpoint in the camera coordinate system.

[0028] Furthermore, in S3, the skeleton of the complete three-dimensional point cloud model of the tobacco shreds is extracted to obtain the centerline curve function; the curvature of the tobacco shreds is calculated based on the centerline curve function, specifically as follows: The centerline curve function r(s) = [x(s), y(s), z(s)] of the complete 3D tobacco point cloud model is generated using a morphological refinement algorithm, where s is the arc length parameter. The arc length L of the tobacco shreds is obtained through the centerline curve function. Based on the centerline curve function r(s), the local curvature is calculated. : ; Based on local curvature Obtaining mean curvature : ; The average curvature of the output reflects the overall curl of the tobacco.

[0029] Example 3: Static tobacco sample testing experiment: (1) Experimental objective: To verify the accuracy and repeatability of the tobacco curl measurement device based on the three-dimensional platform under static conditions. By performing three-dimensional reconstruction and curl calculation on different types of mixed tobacco samples, the accuracy of the device in extracting the spatial curvature features of tobacco under standard conditions is evaluated.

[0030] (2) Experimental conditions: The experiment was conducted in a laboratory with constant temperature (25℃), no wind, and controlled lighting. The aforementioned blue light illumination system (415nm, light intensity approximately 2000lx) and high-resolution binocular industrial camera (2592×1944 pixels, 25mm fixed-focus lens) were used. The XY-axis electric arc stage adjusted the camera posture according to the preset program to sequentially obtain images of tobacco shreds from a top-down view and four sets of tilted angles (5° / 10° / 15° / 20°). Before shooting, the device was calibrated using a checkerboard calibration board to calculate the intrinsic and extrinsic parameters at each angle and obtain the conversion factor from pixels to actual size. The conveyor belt was stationary.

[0031] (3) Sample preparation: Mix ordinary tobacco, cut tobacco, expanded tobacco and reconstituted tobacco in a mass ratio of 50 g: 20 g: 15 g: 15 g, with a total mass of about 200 g. After mixing, randomly scatter the sample in the conveyor belt carrying area to ensure that the sample is evenly distributed and that some tobacco shreds overlap to simulate the situation that may occur in actual production.

[0032] (4) Experimental steps: ① Generate an angle set based on preset parameters: the angle directions are set to the X-axis, Y-axis, and XY-axis, and the angle groups are selected as 5°, 10°, 15°, and 20°; drive the XY-axis electric arc stage to move sequentially to the specified angles in the angle set. After reaching the target angle each time, the blue light illumination system is synchronously triggered to control the binocular industrial camera to acquire images of the tobacco sample at the corresponding angle, ultimately forming a sequence of tobacco images covering multiple perspectives for subsequent 3D reconstruction.

[0033] ② Coordinate calibration is performed on the acquired multi-view tobacco images. The checkerboard calibration method is adopted, and checkerboard images at different angles are captured. The camera rotation matrix R (3×3) and translation vector T (3×1) corresponding to each tobacco image are solved by algorithm to complete the coordinate mapping relationship.

[0034] ③ Based on the image sequence in step 1 and the coordinate mapping relationship in step 2, stereo matching, feature calibration, and point cloud generation and fusion are completed to finally obtain a complete three-dimensional spatial point cloud model of tobacco.

[0035] The stereo matching is based on the depth value of the image at the i-viewpoint calculated using the parallax d of a binocular industrial camera. The formula is: Where f is the focal length of the binocular industrial camera, B is the baseline distance of the binocular industrial camera, and d is the parallax of the binocular industrial camera. The features in the feature calibration specifically include the shape of the tobacco sample and the natural texture of the tobacco sample surface, wherein the natural texture includes fiber orientation and wrinkles; The image with calibrated features at different angles, the corresponding depth values, the mapping relationship between the corresponding camera coordinates and world coordinates are input into the feature point matching algorithm to obtain point cloud data of tobacco sample images at different angles; The point cloud data of the tobacco sample images from different angles were registered using the ICP algorithm: The ICP algorithm uses an error function to minimize the distance of the same feature point under different viewpoints, thus obtaining... , Optimal values ​​improve the registration accuracy of the ICP algorithm: Where n represents all n corresponding feature points. Let be the coordinates of the i-th feature point at any tilted viewpoint in the camera coordinate system. Let be the coordinates of the i-th feature point from the downward viewpoint in the camera coordinate system.

[0036] ④ Perform post-processing and curl calculation on the obtained complete three-dimensional tobacco point cloud model. The complete three-dimensional point cloud model of the tobacco shreds is post-processed and its skeleton extracted to obtain the centerline curve function; the curvature of the tobacco shreds is calculated based on the centerline curve function, the specific process being as follows: Post-processing includes: point cloud filtering and noise reduction to remove noise and isolated points, and point cloud refinement; The centerline curve function r(s) = [x(s), y(s), z(s)] of the complete 3D tobacco point cloud model is generated using a morphological refinement algorithm, where s is the arc length parameter. The arc length L of the tobacco shreds is obtained through the centerline curve function. Based on the centerline curve function r(s), the local curvature is calculated. : ; Based on local curvature Obtaining mean curvature : ; The average curvature of the output reflects the overall curl of the tobacco.

[0037] ⑤ The device records the curling data of each tobacco shred and counts the sample size for subsequent analysis.

[0038] (5) Experimental results: 100 frames of static tobacco images were processed to obtain three-dimensional models and curl measurement results for approximately 2,000 tobacco strands. The results show that the average error of the device in measuring the curl of tobacco strands is approximately ±3.6%, which is lower than that of the traditional two-dimensional projection measurement method; the device can effectively identify and separate overlapping tobacco strands, with an overlapping tobacco strand recognition rate of 89.5%.

[0039] (6) Results: Compared with manual measurement results, the consistency of the curvature and geometric features of the tobacco shreds measured by this device reached over 98%, verifying the measurement accuracy and stability. Through multi-view blue light imaging and three-dimensional skeleton extraction, this device overcomes the problem of low measurement caused by the neglect of tobacco shred overlap and spatial curvature in traditional two-dimensional methods, while providing high-value-added data such as the three-dimensional curvature distribution of tobacco shreds, which is beneficial to the optimization of tobacco processing technology.

[0040] Example 4: Online detection experiment of dynamic blended tobacco: (1) Experimental objective: To verify the robustness, real-time performance, and measurement stability of the three-dimensional curl measurement device under simulated production line conditions. By continuously collecting data under conveyor belt motion, the device's ability to suppress motion blur and illumination fluctuations is evaluated, and the device's online processing efficiency is tested.

[0041] (2) Experimental Apparatus and Samples: A conveyor belt system was used, with the conveyor belt speed set at 0.3 m / s to simulate the transmission speed of an actual tobacco production line. Other equipment, such as the XY-axis electric arc stage 3, blue light illumination system, and binocular industrial camera, had the same parameters as in Example 3. The experimental environment allowed for light intensity fluctuations of ±10%. A tobacco sample (approximately 200 g in total) mixed in the same mass ratio as in Example 3 was used, and data collection began after the conveyor belt had stabilized.

[0042] (3) Experimental steps: ① While the conveyor belt is moving at a constant speed, the camera and blue light illumination device are synchronously triggered at a sampling frequency of 25fps to acquire continuous image sequences from top-down and four sets of tilted viewpoints in various directions. ② Use Time Delay Integration (TDI) mode to synchronize camera exposure with conveyor belt speed to suppress the impact of motion blur on image quality; ③ For the multi-view images acquired at each time point, perform the same three-dimensional reconstruction process as in Example 3: stereo matching to generate point clouds, feature registration and ICP fusion of point clouds, filtering and denoising and extracting skeleton curves; ④ The device calculates the curl value of each tobacco sample in real time, processes 80 frames of images continuously, and automatically counts the curl distribution of each tobacco shred. ⑤ The background uses a multi-threaded approach to optimize the point cloud reconstruction and curvature calculation algorithms to ensure the device's online processing speed under dynamic conditions.

[0043] (4) Experimental Results: During the dynamic online detection process, a total of 80 frames of image data were collected, and approximately 1600 tobacco shreds were detected. The results show that the average curl measurement error of this device under conveyor belt motion conditions is ±4.1%, which is still significantly better than the traditional two-dimensional detection method. The average real-time detection frame rate of the device is approximately 25 fps, which is equivalent to about 0.8 seconds for the curl measurement of each tobacco shred, meeting the detection requirements of 300-500 tobacco shreds per minute. Under the influence of light intensity fluctuations and continuous motion, the measurement accuracy only shows slight changes, indicating that the device has good stability. For overlapping tobacco shred samples, the detection success rate of this device under dynamic conditions exceeds 88%.

[0044] (5) Results Analysis: Experimental results show that the three-dimensional curvature measurement device maintains high accuracy and efficiency under dynamic production line simulation conditions. Compared with traditional two-dimensional methods, this device can provide spatial curvature information of tobacco shreds in real time under motion, and the detection speed is sufficient to meet the requirements of industrial online applications. In addition, dynamic experiments verified the effectiveness of the three-dimensional registration algorithm, ensuring the robustness and reliability of the device.

Claims

1. A device for measuring the curvature of tobacco shreds based on a three-dimensional platform, characterized in that, The application relates to a measurement device for measuring the curling degree of a tobacco sample. The measurement device comprises a conveyor belt conveying mechanism, an illumination system, a camera, a negative pressure adsorption system, an XY-axis electric radian table and a data processing unit. The conveyor belt conveying mechanism is used for conveying a tobacco sample and moving the tobacco sample to a detection area of the conveyor belt conveying mechanism. The XY-axis electric radian table is fixedly connected with the conveyor belt conveying mechanism and is used for realizing angular deviation of the conveyor belt conveying mechanism along X and Y axes. The illumination system is installed above the detection area and is used for illuminating the edge profile of the protruding tobacco sample. The camera is used for obtaining tobacco sample images at different angles through the XY-axis electric radian table. The negative pressure adsorption system is used for ensuring that the tobacco sample is kept stable during transmission and shooting.

2. A three-dimensional stage-based tobacco shankiness measurement apparatus according to claim 1, wherein, The data processing unit is used for image processing of the tobacco sample images at different angles to obtain the curling degree of the tobacco sample.

3. A three-dimensional stage-based tobacco shank curl measurement apparatus according to claim 1, wherein, The illumination system is a blue light illumination system.

4. A three-dimensional stage-based tobacco shank crookedness measuring apparatus according to claim 3, characterized by, The conveyor belt of the conveyor belt conveying mechanism adopts a composite structure, the composite structure comprises a base belt and a breathable layer, a plurality of annular fine slits are uniformly formed on the base belt, the breathable layer is covered on the surface of the base belt, and a plurality of micropores are uniformly distributed on the surface of the breathable layer.

5. A three-dimensional stage-based tobacco shank crookedness measuring apparatus according to claim 3, wherein, The base belt is a black antistatic base material, which effectively reduces optical reflection interference. The breathable layer is a food-grade non-woven fabric, and the micropore aperture is 1-100 mu m.

6. The three-dimensional stage-based tobacco shank crookedness measuring apparatus according to claim 1, wherein, The negative pressure adsorption system comprises a vacuum pump, an air path and an air pressure sensor.

7. An image processing method for a three-dimensional stage-based tobacco shank curl measurement apparatus, the method comprising: capturing an image of a tobacco shank on the three-dimensional stage; and processing the image to determine a curl value of the tobacco shank. The air pressure sensor is used for real-time detection of the air pressure value in the conveyor belt, and when the air pressure value is not within the set threshold value, the vacuum pump is automatically adjusted to maintain the constant adsorption force of the conveyor belt on the tobacco sample. One end of the air path is in communication with the inside of the conveyor belt conveying mechanism, and the other end is in communication with the vacuum pump. The vacuum pump forms a controllable negative pressure field through the air path, the annular fine slits and the uniformly distributed micropores on the breathable layer, so that the tobacco sample is always close to the surface of the conveyor belt during the operation of the conveyor belt conveying mechanism and shooting, and posture deviation caused by air flow disturbance or vibration is prevented. The camera is used for obtaining tobacco sample images at different angles through the XY-axis electric radian table.

8. The image processing method of a three-dimensional stage-based tobacco shank crookedness measuring apparatus according to claim 7, characterized by, The XY-axis electric radian table adjusts the inclination angle of the conveyor belt to generate five groups of perspective images, namely, a top view, two groups of inclined views along the positive and negative directions of the X axis, and two groups of inclined views along the positive and negative directions of the Y axis. The measurement device is realized based on any one of claims 1-6 and comprises the following steps. S1, coordinate calibration is performed on the tobacco sample images at different angles to establish the mapping relationship between a world coordinate system and a camera coordinate system. S2, stereo matching is performed on the tobacco sample images at different angles, feature calibration is performed, then point cloud data of the tobacco sample images at different angles is obtained based on a feature point matching algorithm, and a complete three-dimensional space point cloud model of the tobacco is obtained based on the point cloud data fusion. S3, a center line curve function is obtained by skeleton extraction on the complete three-dimensional tobacco point cloud model, and the curling degree of the tobacco is calculated based on the center line curve function. In S1, the coordinate calibration is specifically performed. A mapping relationship between the world coordinate system and the camera coordinate system is established by using a camera rotation matrix and a translation vector; wherein, , , are camera coordinate system coordinates, , , are world coordinate system coordinates, R is a 3x3 rotation matrix containing camera XYZ three-axis Euler angles, T is a 3x1 translation vector, (R, T) of any image is calculated by using a chessboard calibration method, and a mapping relationship between camera coordinates and world coordinates of any image is established.

9. The image processing method of a three-dimensional stage-based tobacco shank crookedness measuring apparatus according to claim 7, characterized by, In S2, the specific steps of performing stereo matching on the tobacco sample images at different angles, feature calibration, and then fusing point cloud data of the tobacco sample images at different angles based on a feature point matching algorithm are as follows: The stereo matching is based on binocular industrial camera parallax to calculate a depth value of an image at an i-viewing angle , the formula is: ； Wherein, f is the focal length of the binocular industrial camera, B is the baseline distance of the binocular industrial camera, and d is the parallax of the binocular industrial camera. The features in the feature calibration are specifically shapes of the tobacco sample and natural textures on the surface of the tobacco sample, and the natural textures include fiber directions and wrinkles. The images at different angles after feature calibration, corresponding depth values, and mapping relationships between corresponding camera coordinates and world coordinates are input into a feature point matching algorithm to fuse point cloud data of the tobacco sample images at different angles. The point cloud data of the tobacco sample images at different angles is registered through an ICP algorithm: ICP algorithm uses error function to minimize the distance of the same feature points under different perspectives, and gets , optimal value, which improves the registration accuracy of ICP algorithm: ； wherein n is all of the n corresponding feature points, is the coordinate of the i-th feature point under any oblique view angle in the camera coordinate system, is the coordinate of the i-th feature point under the top view angle in the camera coordinate system.

10. The image processing method of a three-dimensional stage-based tobacco shank crookedness measuring apparatus according to claim 7, wherein In S3, a centerline curve function is obtained by performing skeleton extraction on the complete three-dimensional tobacco point cloud model of the tobacco; and the tobacco curling degree is calculated based on the centerline curve function, and the specific process is as follows: A center line curve function r(s)=[x(s),y(s),z(s)] of a complete three-dimensional tobacco shred point cloud model is generated through a morphological refinement algorithm, s is an arc length parameter, and a tobacco arc length L is obtained through the center line curve function; based on the center line curve function r(s), a local curvature : ； Based on local curvature Obtaining average curvature : ； Output average curvature Reflects the overall curl of the tobacco rod.

Images