Analysis of factors affecting measurement accuracy of machine vision inspection equipment

In recent years, machine vision systems have been widely used. They are favored by enterprises for their high precision, high detection efficiency and reduced production costs. Even the best things will have problems. During the use of machine vision systems, due to various factors, the measurement accuracy will be reduced, which will bring inconvenience to enterprises. Today we will talk about the factors that affect the measurement accuracy of machine vision inspection equipment, which can provide some reference when encountering problems.

In some machine vision inspection projects, many customers have encountered problems with inspection accuracy. For example, it is required to measure the external dimensions, inner diameter and outer diameter of the processed parts with an accuracy of 10um. At present, many configurations choose 5 million industrial camera lenses, which can theoretically achieve an accuracy of 10um. However, the actual customer test accuracy is still far from the 10um requirement, and it is very good to achieve 30um. The main reasons for the large error are as follows:

1. Selection of hardware for visual inspection equipment

1. Selection of industrial cameras
Due to the characteristics of the CMOS camera chip itself, when shooting some objects, it is easy to cause poor contrast of the edge contour and large image noise, which will cause errors in software measurement and take more time to solve.

2. Selection of light source
The light source is also a very important part of visual inspection accuracy. In some application scenarios with high requirements, the light generated by the backlight at a certain point will diverge to any angle in space. If circular or cylindrical objects are detected, obvious diffraction will occur at the edge of the measured object, resulting in deviations in the image effect. In addition, the brightness of the light source has a great influence on the brightness of the object.

3. Selection of lens
Because many mechanical parts have height differences and large depth of field. However, ordinary lenses are difficult to shoot due to perspective factors. In the process of software processing, it is difficult to find the most realistic and accurate edge contours, and the image processing algorithm puts forward higher requirements for image processing algorithms.

2. Problems with visual inspection equipment software

1. Software algorithm errors
Even the most rigorous visual inspection methods, calculation formulas and image processing methods, under the influence of different environments and equipment, will inevitably affect the measurement accuracy of the inspection system, causing a certain degree of error, but this error is relatively easy to handle.

2. Calibration error
The calibration process is a necessary process for visual inspection. The system will introduce errors during the calibration process. This method uses multiple images of different positions in the camera's field of view to calibrate standard components, calculates their average value as a correction factor, and eliminates the error caused by lens distortion. But one thing to note is that the calibration process will produce random errors.

3. Imaging system error

The higher the resolution of an industrial camera, the smaller the actual size of the object being measured. The higher the resolution of the imaged object surface, the higher the system detection accuracy. Geometric distortion is a typical system error and an important factor that affects the accuracy of optical detection.

3. Environmental impact in visual inspection

1. Vibration
Vibration is one of the most influential factors in visual inspection. Even slight vibration may cause image blur and distortion, while variable parts may cause different images, and long-term exposure may cause image clarity distortion. Therefore, most industrial cameras are anti-vibration, and it is also necessary to ensure that the camera has good stability during movement to reduce the impact of vibration.

2. Ambient light
Light pollution is difficult to avoid in visual inspection, which will cause the captured image to be distorted and blurred. In visual inspection, high-brightness modulated light sources can be used, while reducing the exposure time of the sensor and reducing the aperture, which can effectively minimize the impact of ambient light.

3. Stains, water vapor and dust Stains
, water vapor and dust are inevitable in some visual inspections. These stains, water vapor and dust will adhere to the camera lens and the surface of the light source to produce occlusion, affecting the quality of imaging and not conducive to visual inspection.

4. Temperature
In general, the temperature of the working environment of industrial cameras is between -5 degrees and 65 degrees. Too high a temperature environment will bring noise to the camera imaging, affecting the quality of imaging, thereby affecting the visual inspection results.

5. Maintenance and care
Visual inspection equipment is very precise, so it is also important to choose the environment in which it is used. You cannot just pick a site casually. A good environment is also good for equipment installation. Whether the visual inspection equipment is in good condition or not is an important prerequisite for the inspection results.

Accuracy of measurement depends on resolution

A decisive factor in providing high accuracy and low uncertainty in machine vision measurements is the resolution of the acquired image. In this context, the term resolution (or image resolution) means the size of a single pixel in real-world units. In simple terms, if a camera sensor contains 1000 pixels in the horizontal direction, and the optics are employed to acquire an image covering an area 1 inch wide in a real-world scene, a single pixel would represent 0.001". Note that this is a fundamental metric that will not be changed by camera manufacturers or analysis software.

How many pixels are enough for a particular application?

As a measure of standard, the smallest unit of measurement in a machine vision system (with some exceptions mentioned later) is a single pixel. As with any measurement system, in order to make repeatable and reliable measurements, it is necessary to use a gauge with a minimum unit of measurement that (as a general rule of thumb) is one-tenth the tolerance band of the desired measurement. In the example just described, it can be estimated that the system provides measurements with an accuracy of approximately +/- 0.005" (0.01" tolerance band, ten times the gauge unit).

Engineers first using machine vision for measurement often severely underestimate the number of pixels required to achieve a desired level of measurement accuracy uncertainty. In fact, it may require multiple cameras, specialized cameras (such as line scan imagers), or multiple views of a single part to achieve the resolution required for a specified inspection tolerance.

Expand the solution if necessary

Sometimes, we can mathematically squeeze extra resolution out of an imaging system using algorithms that report features with sub-pixel repeatability. Some examples are grayscale edge analysis, geometric or correlation searches, regressions such as circle or line fitting, and in some cases connectivity. If sub-pixel results can be considered by using these tools, then the smallest unit of measurement can be smaller than a single pixel, as previously mentioned. Note that sub-pixel capability estimates provided by vendors are just that, estimates, and are typically used for the best imaging, optics, and part representation. Be cautious about using arbitrary sub-pixel expectations as a determining factor in a given system's measurement capabilities. Test the system with actual parts and images to empirically determine sub-pixel capability.

Use high-resolution optics

Imaging is a function of optics and lighting (which we will discuss later, in part). For most applications, the only optics used will be a lens assembly, but the selection of that lens is critical for metrology applications. In addition to providing an image of the proper real-world size to the sensor, for metrology purposes the lens must reproduce the image as accurately as possible without distortion. Additionally, lenses also have a resolution metric, which is typically specified as line pairs per mm or inch (lp/mm, lp/in), and by extension may have a specification for MTF (Modulation Transfer Function) or more simply the ability to produce high contrast at high lp/mm. The higher the pixel count, the more important these lens metrics become. Make sure the optics specified are high-quality, high-resolution products designed for machine vision applications.

Telecentric lenses are very useful for measurement applications in many situations. Telecentric lenses use an optical combination to virtually eliminate all distortion caused by parallax in the image. The result is an image where nearly all of the image is parallel to the sensor. Planar geometric relationships (in the image plane) are fully preserved, making measurements more direct and immediate. As always, test imaging before specification.

For applications requiring very small fields of view (e.g., less than a few millimeters), consider using microscope optics and/or high-magnification optics made specifically for machine vision. These are available from many suppliers. The use of extenders or additional magnifications to push standard optics to higher magnifications is not recommended.

Following thoughts on lighting and part features and presentation.

Choosing the Right Lighting

In metrology, the choice of lighting can play a critical role. Unfortunately, there are no specific rules that can be applied to lighting. Although the physical implementation of automated backlighting on a production line can be a challenge, many metrology applications benefit from backlighting (as described below and partially demonstrated). Front lighting may highlight the edges of features that must be identified for measurement. Consider using low-angle or structured lighting to highlight low-contrast features. When trying to measure very small features (e.g., resolutions below 0.001mm), long wavelength colors such as blue or violet can be used to improve contrast. If the part is in motion (and even if it is not), consider strobing the LED illuminator for optimal intensity and bulb life.