Cognex Vision Systems – The five design elements of a machine vision system

1. Stability of lighting

Industrial vision applications generally fall into four categories: positioning, measurement, inspection, and identification. Among them, the measurement requires the highest photostability, because as only 10-20% of the light changes, the measurement result can deviate by 1-2 pixels. It's not a software issue. This is a change in lighting that causes the edge position of the image to change. Even the most powerful software won't solve the problem. From a system design point of view, it is necessary to exclude ambient light interference while at the same time ensuring an active light source. Luminescence stability. Of course, the increased resolution of hardware cameras is also a way to improve accuracy and resist environmental interference. For example, the space size of the previous camera counterpart was 1 pixel 10 microns, but by increasing the resolution, it became 1 pixel 5 microns. The accuracy rate can be thought to be about doubled, and the disturbance to the environment is naturally enhanced.

2. The position of the workpiece is inconsistent

Generally speaking, whether it is an offline test or an online test, as long as it is a fully automated test equipment, the first step is to be able to find the target to test. Every time the object appears in the field of view, you must be able to know exactly where the object is located. Even if some mechanical fixtures are used, etc., there is no guarantee that the object to be measured will appear in the same position every time. Yes, this requires a positioning feature. If the positioning is not accurate, the position of the measuring tool may not be accurate, and the measurement results can sometimes be quite skewed.

3. Calibration

Generally, high-precision measurements need to be calibrated as follows: first, optical distortion calibration (if a software lens is not used, it is generally necessary to calibrate); Second, the projection distortion calibration, that is, because your installation position error represents the correction of the image distortion and calibration of the image space of the three objects, that is, the specific calculation of the size of the image space corresponding to each pixel.

However, the current calibration algorithms are all based on plane calibration. If the measured physics is not a plane, the calibration needs to be handled by some special algorithms, which cannot be solved by the usual calibration algorithms.

In addition, there are some calibrations. Since no calibration plates are used, special calibration methods must be designed. As a result, the existing calibration algorithms in the software may not fully address the calibration problem.

Fourth, the speed of the object's motion

If the object to be measured is not stationary, but in motion, then we must consider the effect of motion blur on the accuracy of the image (blurry pixels = object movement speed * camera exposure time), which cannot be solved by software.

5. The measurement accuracy of the software

In measurement applications, the accuracy of the software can only consider 1/2-1/4 pixels, preferably 1/2, and cannot achieve the accuracy of 1/10-1/30 pixels like in positioning applications, because in measurement applications the software can extract very few feature points from the image.

Machine vision's motion speed and measurement accuracy play an important role in the entire product. The speed of movement is inversely proportional to the detection ability. The faster the movement, the worse the quality of the inspection. Therefore, it is important to improve the accuracy of the motion and the detection of details. Moreover, machine vision is easy to realize information integration, which is the basic technology to realize computer integrated manufacturing.

Cognex Vision Systems – The five design elements of a machine vision system

Cognex Vision Systems – The five design elements of a machine vision system

Sections viewed

Contact

working hours