Abstract: Machine vision replaces human eyes in the identification, judgment, and measurement of target objects. In industrial production, compared to traditional measurement and inspection methods, the greatest advantages of machine vision technology are its speed, accuracy, reliability, and intelligence. Machine vision plays an irreplaceable role in improving product inspection consistency, product manufacturing safety, reducing worker labor intensity, and achieving efficient, safe production and automated management in enterprises. This paper introduces a method to improve the accuracy of image pattern recognition by optimizing the illumination of the vision system.
Foreword
With the development of science and technology, the application fields of industrial robots are constantly expanding. Currently, industrial robots are not only used in traditional manufacturing industries such as mining, metallurgy, petroleum, chemicals, and shipbuilding, but also in high-tech fields such as nuclear energy, aviation, aerospace, medicine, and biochemistry, as well as service industries such as home cleaning and medical rehabilitation. Machine vision, as a core technology of industrial robots, mainly studies how computers simulate human visual functions to identify, judge, and measure target objects.
The core of a machine vision system is image acquisition and processing. All information originates from the target image, and the quality of the target image itself is crucial to the entire vision system. The light source is a significant factor affecting the image quality of a machine vision system, as it directly impacts the quality of the input data and at least 30% of the application's effectiveness. By optimizing the lighting appropriately, the target information in the image can be optimally separated from the background information, significantly reducing the difficulty of image processing algorithms for segmentation and recognition, improving the system's positioning and measurement accuracy, and ultimately enhancing the system's reliability and overall performance.
Currently, there are no universally applicable machine vision lighting devices on the market. Existing lighting cannot be tailored to every specific application. Only through analysis and optimization of the specific lighting device can the best imaging effect be achieved. This is precisely the value of the light source in a machine vision system. Optimizing the lighting of a machine vision system can effectively improve the quality and effect of image recognition and increase the working efficiency of industrial robots.
This paper focuses on industrial robot chess teaching instruments. Addressing the issue of low accuracy in chess piece recognition by industrial robot vision systems, it proposes an optimization scheme for the lighting of robot vision systems. By optimizing the lighting devices in the system, the recognition accuracy of the vision system can be improved.
1. Composition of a machine vision system
A typical machine vision system can be divided into three parts: image acquisition, image processing, and motion control. Its main components include a machine vision light source, optical lens, industrial camera, sensors, image analysis and processing software, and communication interfaces.
(1) Light source
In current machine vision applications, high-quality light sources and illumination schemes are often crucial to the success of the entire system. The coordination between the light source and illumination scheme should highlight the object's features as much as possible, creating a clear distinction between the parts of the object that need to be detected and those that are not important. Among these, LED light sources are increasingly being used in modern machine vision systems due to their numerous advantages.
(2) Optical lens
Optical lenses are analogous to the lens of the human eye and are crucial in machine vision systems. Key performance indicators of a lens include focal length, aperture ratio, magnification, and interface.
(3) Camera
The camera is the most important component of a machine vision system for acquiring raw information. Currently, the main types used are CMOS cameras and CCD cameras. CCD cameras are widely used in both commercial and industrial fields due to their compact size, reliability, and high resolution.
(4) Image acquisition card
In PC-based machine vision systems, the image acquisition card is a crucial device for controlling the camera to take pictures, completing image acquisition and digitization, and coordinating the entire system.
(5) Visual sensor
The modular components of the sensor integrate a light source, camera, image processor, and standard control and communication interfaces, forming an intelligent image acquisition and processing unit. The internal program memory can store image processing algorithms, and various algorithms can be programmed and downloaded to the vision sensor's program memory using a PC and dedicated configuration software.
2. Light source technology
The coordination of light sources and illumination schemes should highlight the object's feature parameters as much as possible, increasing image contrast while ensuring sufficient overall brightness; changes in object position should not affect image quality. The choice of light source must conform to the required geometry, illumination brightness, uniformity, etc. A good light source can improve the overall system resolution and reduce the burden on subsequent image processing. An unsuitable light source can cause many problems for the machine vision system; for example, camera artifacts and overexposure can hide much important information; shadows can cause false edge detection; reduced signal-to-noise ratio and uneven illumination can increase the difficulty of selecting image processing thresholds. To ensure good images, it is essential to choose a suitable light source. The main types of light sources in machine vision are as follows:
2.1 Front Light Source
A front light source refers to a light source placed in front of the object being measured. This type of illumination is called "front-lit illumination," as shown in Figure 1. Front light sources can be further divided into "high-angle" and "low-angle" types, the difference being the angle between the light source and the surface of the object being measured. When selecting "high-angle illumination" or "low-angle illumination," the mechanism of the part of the object being measured should be considered.
Figure 1 Schematic diagram of front light source illumination
2.2 Backlight
The backlight is positioned opposite to the front light, behind the object being tested, as shown in Figure 2. By illuminating the object with the backlight, a shadow is cast relative to the camera, allowing observation of the interior of an opaque object or the transparent object itself. This clarifies the edges of the translucent and opaque parts of the object, laying the foundation for image edge extraction. It is primarily used for contour detection of objects, defect detection of transparent bodies, LCD text inspection, size and shape inspection of small electronic components, appearance and size inspection of bearings, and appearance and size inspection of semiconductor lead frames.
Figure 2 Schematic diagram of rear light source illumination
2.3 Ring Light Source
A ring light source provides large-area, uniform illumination for the object being measured. In practical applications, the ring light source is coaxially mounted with the CCD lens, typically aligned with the lens edge. The advantage of a ring light source is that it can be directly mounted on the lens, as shown in Figure 3. When the distance from the object is appropriate, it can significantly reduce shadows, improve contrast, and achieve large-area illumination. However, if the distance is inappropriate, it can cause ring-shaped reflections.
Figure 3 Schematic diagram of ring light source illumination
The Mitsubishi RV-2F industrial robot chess teaching instrument mainly consists of a Mitsubishi RV-2F industrial robot, a Mitsubishi FX3U series PLC, a Mitsubishi GOT1000 series touch screen, a Mitsubishi MR-J4 series servo motor, a Cognex industrial camera, and a Banner safety light curtain. It can perform actions such as grasping, placing, and recognizing chess pieces on a standard Chinese chess board, as well as moving and capturing pieces. The vision system lighting in this system uses a ring-shaped illumination source, as shown in Figure 4.
Figure 4. Mitsubishi RV-2F industrial robot chess teaching instrument
Figure 5. Circular lighting source
3. Analysis of the Ring Light Source for Industrial Robot Chess Teaching Instruments
The initial ring-shaped lighting source used horizontally arranged LEDs, which could provide relatively uniform illumination to the entire workpiece. Its lighting principle is shown in Figure 5.
Figure 6. Schematic diagram of a purely horizontally arranged ring-shaped lighting source.
This ring-shaped illumination source mounts LEDs onto a planar structure. Its simple design and easy assembly result in low cost. However, its biggest drawback is uneven brightness at the center of the field of view. Images acquired by the CCD exhibit a low brightness at the center and high brightness around the perimeter. This illumination is also difficult to match well with the aperture size of the COGNEX1100 industrial vision sensor's imaging lens, causing the characteristic details of the acquired image to be masked, posing significant challenges to image capture and discrimination by In-SightExplorer vision.
To achieve a more uniform lighting source, the arrangement of LED particles in Figure 5 can be optimized. An aspherical structure is adopted, embedding the LED particles on the aspherical surface, ultimately providing illumination to the workpiece through multi-angle diffusion. The lighting principle is shown in Figure 6.
Figure 7 Schematic diagram of aspherical ring lighting source
Compared to existing ordinary LED ring light sources, the aspherical ring light source shown in Figure 6 has the advantage of producing a uniform and evenly distributed lighting effect. A comparison of the two is shown in Figures 7 and 8. A comparison of Figures 7 and 8 reveals that the latter provides more uniform and softer lighting.
Figure 8. Reflection diagram of a purely horizontally arranged ring-shaped lighting source.
Figure 9. Reflection diagram of aspherical array lighting source
4. Analysis of the Ring Light Source for Industrial Robot Chess Teaching Instruments
The In-SightExplorer spreadsheet view is the programming interface for Cognex In-SightMicro series industrial vision sensors, as shown in Figure 9. The In-SightExplorer spreadsheet view is as easy to use as Excel; simply drag and drop the required commands into the appropriate positions on the spreadsheet and edit the cell data. The Mitsubishi RV-2F industrial robot chess vision system uses the image comparison command FindPatterns, as shown in Figure 10. It simply takes a model of the image, compares the results, calculates the similarity value, performs the corresponding four-arithmetic-decimal operations, and sends the binary number to the network.
As shown in Figure 9, the exposure time of a typical camera only needs to be adjusted to around 2-3 seconds. ERR in the figure indicates an error. The total score is 100 points, and a score of 60-70 is generally sufficient. The score is determined by comparing image features. When the score is greater than the score threshold, the value will be displayed as 1; when it fails the score threshold, the value will be displayed as 0. 1 indicates successful image recognition, and 0 indicates no image recognition or failure.
Figure 10 In-Sight Explorer Spreadsheet View
Figure 11 Image sampling interface
An optimized light source was installed in the industrial robot chess teaching system. Images of chess pieces of the same type were selected and photographed under different ring-shaped lighting sources to obtain images, as shown in Figures 11 and 12 below.
Figure 12 Reflection diagram of a purely horizontally arranged ring-shaped illumination source
Figure 13 Reflection diagram of aspherical array lighting source
Image comparison reveals that the image in Figure 12 has higher brightness and better uniformity. After image processing and system comparison, Figure 12 successfully passed image recognition, while Figure 11 showed a recognition error.
5. Conclusion
This paper modifies the lighting of the chess vision system of the Mitsubishi RV-2F industrial robot, optimizing the common horizontally arranged LED ring light source into a non-spherical arranged LED ring light source, which effectively improves the brightness and uniformity of the target object's illumination and increases the image recognition success rate of the vision system.
For more information, please follow the Machine Vision channel.
