English is currently available.
Machine Vision DISCUSSION
We want to read printed codes and check label position on boxes moving at 0.5 m/s on a conveyor. The field of view is 200 mm wide with a 1920 × 1200 camera. A low-cost rolling shutter camera gave images where the boxes look skewed and the print is smeared, even though the picture looked bright enough.
How do I calculate the exposure time needed to freeze the motion, and is a global shutter camera required for this?
Blur is the distance the object travels during the exposure, expressed in pixels. Your scale is 200 mm / 1920 pixels = 0.104 mm per pixel. At 500 mm/s the box moves 0.104 mm in 0.104 / 500 s = 208 µs. So to keep the blur to one pixel the exposure must be about 200 µs or shorter. For code reading a pixel or two is usually tolerable, while for edge measurement you want well under one.
For comparison, an automatic exposure of 10 ms in indoor light means 5 mm of travel, which is 48 pixels of smear. That explains the unreadable print.
The skew is a separate effect from blur. A rolling shutter sensor exposes and reads its rows one after another, so the top and bottom of the image are captured at different times. If reading out 1200 rows takes 16 ms, the box moves 8 mm, about 77 pixels, between the first and the last row, and vertical edges become slanted. Shortening the exposure does not remove this, because the row-to-row delay stays the same.
A global shutter sensor exposes all pixels at the same instant, so there is no geometric distortion, only the blur set by the exposure time. For moving parts it is the standard choice.
A 200 µs exposure needs much more light than room lighting provides: compared with 10 ms it collects 50 times less. The usual answer is a strobed LED light synchronised to the camera. With a suitable strobe controller that limits pulse length and duty cycle, LED lights can be driven well above their continuous rating for short pulses, and the camera's strobe output times the pulse to the exposure. Opening the aperture and raising the gain also help, at the cost of depth of field and noise respectively.
If the ambient light is weak by comparison, the effective exposure becomes the length of the light pulse.
Trigger the camera from the process rather than letting it free-run. A photoelectric sensor on the conveyor fires a hardware trigger so each box is captured once, at the same position in the image. At 0.5 m/s a trigger timing uncertainty of 1 ms shifts the box by 0.5 mm, about 5 pixels, so use the camera's hardware trigger input rather than a software trigger over USB or Ethernet, whose latency varies.
If the conveyor speed is not constant, an encoder on the belt gives a position-based delay instead of a time-based one.
Back to Machine Vision community
