Line structured light online thickness measurement system and method
Through the linear structure light online thickness measurement method combined with semantic segmentation model and light hood, the problem of high ambient light impact and image quality requirements in traditional methods is solved, and higher accuracy measurement is achieved, especially in complex scenarios degree identification.
Patent Information
- Application Number
- CN202411478983.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2024-12-13
AI Technical Summary
Traditional linear structured light online thickness measurement methods have problems such as ambient light influence, high image quality requirements, difficulty in processing complex or dynamic scenes, and low measurement accuracy of objects made of highly reflective or transparent materials.
The semantic segmentation model is used to combine the light hood and a high-speed camera to generate linear structured light through the laser emitter and optical system. After capturing and performing grayscale processing, the center line of the light bar is extracted using a deep learning network to reduce the impact of ambient light and improve measurement accuracy.
The semantic center method reduces the impact of surface characteristics, improves measurement accuracy, reduces the requirements for image quality, can handle complex scenes and maintain high accuracy.
Smart Images
Figure CN119124014A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of thickness measurement, and in particular relates to a line structured light online thickness measurement method and system. Background Art
[0002] Online measurement technology plays a very important role in the industrial industry, and its results can directly guide product quality control and production process optimization. Although traditional online measurement methods, such as sensor technology and machine vision technology, have achieved online real-time measurement to a certain extent, they still have some problems such as insufficient accuracy, poor environmental adaptability, and high cost.
[0003] In recent years, with the development of computer vision and deep learning technology, the online measurement method based on line structured light has gradually shown its advantages. This method uses a high-speed camera to capture the reflection image of the line structured light on the surface of the object, reconstructs the three-dimensional spatial position of the center line of the structured light through the two-dimensional image, and calculates the thickness of the object, thereby achieving high-precision measurement. However, there are the following defects:
[0004] 1. Lack of arrangement to limit the influence of ambient light. Excessive, dark, and changing ambient light may affect the effect of adaptive threshold segmentation processing, resulting in inaccurate probability map of light band pixels;
[0005] 2. Over-reliance on precise pixel coordinate operations, requiring higher image quality;
[0006] 3. Adaptive threshold segmentation of point clouds cannot handle complex or dynamically changing scenes;
[0007] 4. This method is somewhat dependent on the reflective properties of the object surface. For objects with highly reflective or transparent materials, it may be difficult to obtain a cleaned stripe image, which affects the measurement accuracy.
[0008] Even using the same image processing approach, the center extraction of the laser line is directly affected by the visual appearance of the target, which is mainly affected by the relative position between the laser line and the target, the relative pose between the camera and the laser line, the resolution of the image taken by the camera, and the unpredictable visual appearance changes when measuring targets with new materials or new shapes. Summary of the invention
[0009] The present invention provides a line structured light online thickness measurement method and system, and the technical problem to be solved is: reducing the influence caused by surface characteristics under a single scale, thereby improving the measurement accuracy.
[0010] In order to solve the above technical problems, the present invention provides a line structured light online thickness measurement system, which is characterized by: comprising a semantic segmentation model, a light shield and a high-speed camera arranged in the light shield, a laser transmitter, an optical system, and a heat dissipation device;
[0011] The laser transmitter and optical system are located above the object to be measured, and the heat dissipation device is used to ensure the constant surface temperature of the object to be measured;
[0012] The laser transmitter emits a laser light source to the optical system, which converts the laser into a line structured light band through the optical system. When the line structured light band continues to propagate from the optical system to the surface of the object to be measured, specular reflection and diffuse reflection occur. The reflected image at this time is captured by a high-speed camera, and the image is gray-processed to obtain a gray-scale image, thereby enhancing the interpretability of the image.
[0013] The semantic segmentation model obtains the thickness data of the object by analyzing the position changes of the light bars in the captured grayscale image.
[0014] A line structured light online thickness measurement method, characterized by comprising the following steps:
[0015] S1. The laser transmitter emits a laser light source to the optical system, and the optical system converts the laser into a line structured light band. When the line structured light band continues to propagate from the optical system to the surface of the object to be measured, specular reflection and diffuse reflection occur. The reflected image at this time is captured by a high-speed camera, and the image is gray-processed to enhance the interpretability of the image.
[0016] S2. Process the obtained grayscale image through the semantic segmentation model to enhance the semantic part of the center line:
[0017] Use the trained semantic segmentation model to output a probability map based on the received grayscale image, in which each pixel captured by the high-speed camera is marked as belonging to the upper, middle and lower line segments. The upper line segment is formed by the reflection of light from the convex part of the object surface, the lower line segment is formed by the reflection of light from the concave part of the object surface, and the middle line segment is formed by the reflection of light from the flat part of the object surface. Threshold the grayscale image according to the probability map, and generate a new image after fusion that contains both grayscale information and probability information, thereby enhancing the semantic part of the center line of the light strip.
[0018] S3. Find the center line of the three line segments:
[0019] The semantic segmentation model performs a weighted average of the brightness values and positions of all pixels to obtain the center position of each column of light stripes, thereby extracting the center line position information of the light stripes of the entire image;
[0020] S4. Finally, the distance between the endpoint of the midline segment and other adjacent segments is calculated and used as the predicted thickness of the cube target.
[0021] Beneficial effects: The present invention uses image processing based on the semantic center method and computer vision algorithm to identify and extract specific features in an image, which can reduce the impact caused by surface characteristics at a single scale, thereby improving measurement accuracy.
[0022] By utilizing the deep learning network, the model is trained to extract features from a large number of line structured light images. At the same time, the multi-scale and multi-feature fusion learning characteristics of the large model enable the trained model to capture features at different resolutions, and perform information processing similar to "context judgment" based on the previously and subsequently related pixel information, further reducing the interference caused by surface characteristics.
[0023] The semantic center method is usually used to determine the category of the collected pictures. The present invention creatively applies the semantic center method to the online thickness measurement of line structured light, interprets and determines the positions of all bright pixels of the line structured light strips, thereby obtaining the position of the center line of the line structured light. By studying the deformation position changes of the light strips in different thickness parts of the object, the overall requirements of the line structured light measurement technology for image quality are reduced, and more accurate measurement data are output. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 Schematic diagram of thickness measurement using line structured light based on semantic center method;
[0025] Figure 2 Schematic diagram of the line structured light online thickness measurement system. DETAILED DESCRIPTION
[0026] In order to make the purpose, content and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below.
[0027] The invention proposes a line structured light online thickness measurement system, comprising a light shield and a high-speed camera arranged in the light shield, a laser transmitter, an optical system, a semantic segmentation model, and a heat dissipation device;
[0028] The high-speed camera uses a CCD camera, and the optical system uses a grating structure.
[0029] The laser emitter and the optical system are perpendicular to the surface of the object to be measured and are located above the object to be measured. The heat dissipation device is used to ensure the constant surface temperature of the object to be measured.
[0030] The laser transmitter emits a laser light source to the optical system, which converts the laser into a line structured light band through the optical system. When the line structured light band continues to propagate from the optical system to the surface of the object to be measured, specular reflection and diffuse reflection occur. The reflected image at this time is captured by a high-speed camera, and the image is gray-processed to obtain a gray-scale image, thereby enhancing the interpretability of the image.
[0031] In grayscale images, due to the different heights of the surface, light strips will be blocked and reflected to varying degrees, causing changes in the image that should have been flat. The semantic segmentation model obtains the thickness data of the object by analyzing the position changes of the light strips on the captured grayscale image.
[0032] The semantic segmentation model uses a deep learning network to train the model to extract features from a large number of line structured light images. At the same time, the multi-scale and multi-feature fusion learning characteristics of the large model enable the trained model to capture features at different resolutions, and perform information processing similar to "context judgment" based on the previously and subsequently related pixel information, further reducing the interference caused by surface characteristics.
[0033] It is necessary to reduce the impact of ambient light on the measurement results in an environment restricted by a light shield.
[0034] A line structured light online thickness measurement method comprises the following steps:
[0035] S1. The laser transmitter emits a laser light source to the optical system, and the optical system converts the laser into a line structured light band. When the line structured light band continues to propagate from the optical system to the surface of the object to be measured, specular reflection and diffuse reflection occur. The reflected image at this time is captured by a high-speed camera, and the image is gray-processed to enhance the interpretability of the image. In the captured gray-scale image, due to the different heights of the surface, the light strip will be blocked and reflected to different degrees, thereby changing the image that should have been flat. By analyzing the position change of the light strip on the image, the thickness data of the object is obtained.
[0036] S2. Process the obtained grayscale image through the semantic segmentation model to enhance the semantic part of the center line:
[0037] At this time, the trained semantic segmentation model is used to output a probability map based on the received grayscale image, in which each pixel captured by the high-speed camera is marked as belonging to the upper, middle and lower line segments. The upper line segment is the line segment formed by the reflection of light from the convex part of the object surface, the lower line segment is the line segment formed by the reflection of light from the concave part of the object surface, and the middle line segment is the line segment formed by the reflection of light from the flat part of the object surface. The grayscale image is thresholded according to the probability map, and the new image generated after fusion contains both grayscale information and probability information, thereby enhancing the semantic part of the center line of the light strip.
[0038] S3. Find the center line of the three line segments:
[0039] The semantic segmentation model performs a weighted average of the brightness values and positions of all pixels to obtain the center position of each column of light stripes, thereby extracting the center line position information of the light stripes of the entire image;
[0040] S4. Finally, the distance between the endpoint of the midline segment and other adjacent segments is calculated and used as the predicted thickness of the cube target.
[0041] The heat sink device ensures that the temperature of the object surface is constant during laser irradiation.
[0042] Before the specific measurement process, the high-speed camera needs to be calibrated, such as the focal length, principal point, and the fixed position of the camera relative to the world coordinate system; the position of the laser line in three-dimensional space is reconstructed through the laser plane equation.
[0043] The present invention uses a light shield for online measurement to effectively block external ambient light, reduce the impact of natural light on the quality of captured images, and improve the signal-to-noise ratio of the image. The active heat dissipation device heat dissipation technology keeps the device at the optimal operating temperature at all times, avoiding thermal stability problems caused by long-term operation of the system or objects being exposed to strong light.
[0044] The use of advanced artificial intelligence technology and the introduction of a deep learning semantic segmentation model can accurately identify and extract the line structured light centerline in the structured light image. Even in complex situations such as rough object surfaces or severe mirror reflections, the high accuracy of centerline recognition and extraction can be maintained.
[0045] The advantage of the present invention is that, under the condition of ensuring a constant measurement environment, sub-pixel level light band center extraction is achieved through a semantic segmentation model. Compared with relying on a point cloud image processing algorithm, the structured light center line can be extracted more quickly and accurately, thereby ensuring efficient and accurate online measurement.
Claims
1. A line structured light online thickness measurement system, characterized by: It includes a semantic segmentation model, a light shield, a high-speed camera arranged in the light shield, a laser transmitter, an optical system, and a heat dissipation device; The laser transmitter and optical system are located above the object to be measured, and the heat dissipation device is used to ensure the constant surface temperature of the object to be measured; The laser transmitter emits a laser light source to the optical system, which converts the laser into a line structured light band through the optical system. When the line structured light band continues to propagate from the optical system to the surface of the object to be measured, specular reflection and diffuse reflection occur. The reflected image at this time is captured by a high-speed camera, and the image is gray-processed to obtain a gray-scale image, thereby enhancing the interpretability of the image. The semantic segmentation model obtains the thickness data of the object by analyzing the position changes of the light bars in the captured grayscale image.
2. The line structured light online thickness measurement system according to claim 1, characterized in that: The high-speed camera uses a CCD camera.
3. The line structured light online thickness measurement system according to claim 1, characterized in that: The optical system adopts a grating structure.
4. The line structured light online thickness measurement system according to claim 1, characterized in that: The laser emitter and the optical system are perpendicular to the surface of the object to be measured.
5. The line structured light online thickness measurement system according to claim 1, characterized in that: The semantic segmentation model uses a deep learning network to train the model to extract features from a large number of line structured light images. At the same time, the multi-scale and multi-feature fusion learning characteristics of the large model enable the trained model to capture features at different resolutions, and perform information processing similar to "context judgment" based on the previously and subsequently related pixel information, further reducing the interference caused by surface characteristics.
6. The line structured light online thickness measurement system according to claim 1, characterized in that: The lens hood can effectively block external ambient light, reduce the impact of natural light on the quality of captured images, and improve the signal-to-noise ratio of the image.
7. The line structured light online thickness measurement system according to claim 1, characterized in that: The heat dissipation device keeps the equipment at the optimal operating temperature at all times to avoid thermal stability problems caused by long-term operation of the system or objects being exposed to strong light.
8. The line structured light online thickness measurement system according to claim 1, characterized in that: Sub-pixel light band center extraction is achieved through semantic segmentation model.
9. A line structured light online thickness measurement method, characterized in that: The steps are as follows: S1. The laser transmitter emits a laser light source to the optical system, and the optical system converts the laser into a line structured light band. When the line structured light band continues to propagate from the optical system to the surface of the object to be measured, specular reflection and diffuse reflection occur. The reflected image at this time is captured by a high-speed camera, and the image is gray-processed to enhance the interpretability of the image. S2. Process the obtained grayscale image through the semantic segmentation model to enhance the semantic part of the center line: Use the trained semantic segmentation model to output a probability map based on the received grayscale image, in which each pixel captured by the high-speed camera is marked with the probability of belonging to the upper, middle and lower line segments. The upper line segment is the line segment formed by the reflection of light from the convex part of the object surface, the lower line segment is the line segment formed by the reflection of light from the concave part of the object surface, and the middle line segment is the line segment formed by the reflection of light from the flat part of the object surface. The grayscale image is thresholded according to the probability map, and the new image generated after fusion contains both grayscale information and probability information, thereby enhancing the semantic part of the center line of the light strip; S3. Find the center line of the three line segments: The semantic segmentation model performs a weighted average of the brightness values and positions of all pixels to obtain the center position of each column of light stripes, thereby extracting the center line position information of the light stripes of the entire image; S4. Finally, the distance between the endpoint of the midline segment and other adjacent segments is calculated and used as the predicted thickness of the cube target.
10. The method for measuring thickness by line structured light online according to claim 9, characterized in that: Before the specific measurement process, the high-speed camera needs to be calibrated, including the focal length, principal point, and the fixed position of the camera relative to the world coordinate system; the position of the laser line in three-dimensional space is reconstructed through the laser plane equation.