What are the difficulties in the design of machine vision systems?

First: The Stability of Lighting

In industrial vision systems, applications are typically categorized into four main areas: positioning, measurement, inspection, and identification. Among these, measurement requires the highest level of lighting stability. Even a small change in illumination—such as 10-20%—can cause a deviation of 1-2 pixels in the measurement results. This is not a software issue but rather a physical one, as changes in lighting affect the edge positions in the image. No matter how advanced the software is, this problem must be addressed at the system level. To ensure accurate measurements, ambient light interference must be minimized, and a stable, active light source should be used. Additionally, improving the resolution of the camera can also enhance accuracy and reduce environmental interference. For instance, if a previous camera had a pixel size of 10 micrometers per pixel, upgrading to a higher-resolution camera that provides 5 micrometers per pixel can effectively double the accuracy, making the system more robust against external factors.

Second: Inconsistency in Workpiece Position

Whether it's offline or online testing, automated systems must first locate the target object. Even with mechanical fixtures, it’s difficult to guarantee that the object will always appear in the exact same position within the field of view. This is why precise positioning is essential. If the positioning is inaccurate, the measuring tool might not align correctly, leading to significant measurement errors. Proper alignment ensures consistent and reliable results across multiple tests.

Third: Calibration

High-precision measurement systems often require several types of calibration. These include optical distortion calibration (if a software lens isn’t used), projection distortion calibration (to correct for installation errors), and object space calibration, which calculates the real-world size corresponding to each pixel. However, most current calibration algorithms are based on planar surfaces. If the object being measured is not flat, special algorithms may be needed, as standard methods won’t work. Additionally, some systems don't use calibration plates, requiring custom calibration techniques that go beyond standard software solutions.

Fourth: Object Movement Speed

When the object being measured is moving, motion blur becomes a critical factor affecting image quality. The amount of blur depends on the object’s speed and the camera’s exposure time. This is a hardware limitation that cannot be fully corrected by software alone. Ensuring proper lighting and using faster cameras can help reduce the impact of motion blur, especially in high-speed applications.

Fifth: Software Measurement Accuracy

In measurement applications, the software can typically achieve an accuracy of around 1/2 to 1/4 of a pixel, ideally 1/2. This is much less precise than in positioning applications, where sub-pixel accuracy (1/10 to 1/30 of a pixel) is common. This difference occurs because measurement software usually extracts only a few key feature points from the image, limiting its ability to detect fine details. Therefore, hardware improvements and careful system design are crucial for achieving high-accuracy measurements in real-world environments.