Sunday, March 28, 2010

GigE – the future of machine vision?

Ned Lecky is excited about GigE cameras. Very excited. In fact he thinks they are going to take over the world. Well not the whole world, obviously, but definitely the machine vision world, or so he says in “GigE Camera Myths Debunked,” (Vision & Sensors, March 2010.)

I like GigE Cameras. I find them easy to work with, and I do agree that Power over Ethernet (PoE) will make GigE about as versatile as FireWire, but I’m not sure they’ll eliminate FireWire as Ned suggests.

I’m not an out-and-out techie but from my reading, particularly “
GigE Vision – CPU Load and Latency” (White Paper available for download from the Basler website,) there are some downsides. These seem to fall into two areas: CPU load and latency/jitter.

Starting with CPU load, this appears to be related to the number of packets transmitted by the camera, rather than the size of the packets. In other words, it might not be beneficial to use a VGA GigE camera rather than a similar VGA FireWire product since the GigE cam will gobble up more CPU resource. However, when you need to work with bigger images, say 5Mp, the greater bandwidth of GigE, combined with the fact that the CPU load doesn’t grow, seem to give it a significant advantage.

On the latency and jitter front, the data provided by Basler seems to show that. For a software-triggered camera, GigE has both greater latency and greater jitter (read the paper to see the actual numbers.) We’re only talking about a difference in latency of 300 microseconds, so in many applications that may not matter, but in a high-speed motion application it could be important.

So to Ned’s point that GigE cameras (will) rule; I see Firewire sticking around for lower resolution applications – say 2Mp and smaller – where the ease of 1394a is attractive, and GigE dominating in the higher res, higher speed arena where it will of course come up against CameraLink.

1 comment:

Anonymous said...

Thanks for the great comments on my article. I look at GigE vs. CameraLink as a more valid engineering trade-off than GigE vs. FireWire, though. In our company, we have found the FireWire cable length restriction to render it unsuitable for so many machine vision applications that we prefer to train our programmers just on GigE, which always works. We also find GigE to be more reliable and robust in multiple camera applications than FireWire or USB.