How the configurator works
The lens is a thin lens with its principal plane at the working distance \(s\) from the part. The sensor of width \(w\) and height \(h\) must see a field of at least \(W \times H\), so the magnification cannot exceed
In automatic mode it picks the longest standard focal length at or below \(f_\text{ideal}\) (4, 5, 6, 8, 12, 16, 25, 35, 50, 75, or 100 mm), so the field is always covered. The actual magnification is then \(m = f/(s - f)\), each pixel of pitch \(p\) covers \(p/m\) on the part, and the smallest feature spans \(d\,m/p\) pixels.
Depth of field, diffraction, and motion
Depth of field uses the exact thin-lens limits with the acceptable blur \(c\) (in pixels) and hyperfocal distance \(H = f^2/(Nc) + f\), the same model as the lens selection calculator. The Airy disk at the sensor is \(2.44\,\lambda N(1+m)\), and the ideal-lens MTF at the sensor Nyquist frequency \(1/(2p)\) shows whether the optics or the pixels limit resolution.
Motion blur in pixels is the distance the part moves during the exposure divided by the pixel size on the part, \(v\,t_\text{exp}\,m/p\). Keep it under one pixel for measurement and under about two for detection. Rolling-shutter sensors also skew moving parts, because rows expose at different times.
Bandwidth and lens format
The data rate is pixels per frame × bytes per pixel × frame rate. The interface limits are practical sustained payloads after protocol overhead: about 115 MB/s for GigE, 380 MB/s for USB3, 1.15 GB/s for 10GigE, and 1.2 GB/s per CXP-12 link. The sensor's own maximum frame rate is not modeled, so check the camera datasheet as well.
A lens must project an image circle at least as large as the sensor diagonal. The configurator suggests the smallest standard lens format that covers the sensor and the usual mount for that format. C-mount covers sensors up to about 1.1″ (17.6 mm diagonal) with lenses rated for it; larger sensors need TFL-II, M42, F-mount, or M58 lenses.
Lighting
Lighting decides contrast, and contrast decides whether a feature that spans enough pixels is actually detectable. The suggestions follow standard practice: backlights for silhouettes and gauging, coaxial light for flat reflective parts, dome light for curved shiny parts, low-angle dark field for scratches and particles, and patterned backgrounds for transparent parts. The machine vision optics article covers the geometry in more depth.
Assumptions and limits
- Thin-lens, paraxial model. Real lenses have separated principal planes and a minimum object distance, so confirm the working distance with the lens maker's data or a test image.
- Telecentric lenses do not follow \(m = f/(s - f)\). For gauging with height variation, the perspective check suggests when a telecentric lens is worth it.
- Color cameras with a Bayer filter resolve fine detail at roughly half the monochrome pixel density. Use monochrome unless the decision depends on color.
- Sensor data are the maker's full active arrays. Effective resolution, frame rate, and noise depend on the camera built around the sensor.