Machine vision


Thread Starter


How can I get the image of an object, identify that object, and find its co ordinates? Which camera and image processing software will fit this requirement?
We have used the DVT camera for this type of thing to pick up parts randomly oriented on a conveyor with a pick and place robot. Pretty easy once you spend a few hours with the software (free).

The camera, lighting and cables were about $5k (we went with a higher end camera that would allow 100+ parts a minute pick rates).

Basic procedure is to create virtual sensor(s) and chain together various canned functions to evaluate the data that you want from the part(s).

This involves photographing the part with the camera, drawing a box (or some other shape) around the item of interest in the picture and then identifying key surfaces, edges, colors, etc. and assigning rules to these artifacts.

Each identification task you assign is a "sensor" - all of this is done on your PC. Once you are done with your sensors, download them to the camera and you are ready to go.

Typically you have Ethernet comms with modbus TCP/IP or you can do serial to a number of different protocols or manage the whole thing with digital I/O.

I don't work for or sell DVT... just integrated a project recently.

Ken Brown
Applied Motion Systems, Inc.

Curt Wuollet

I use standard NTSC (or PAL) video cameras, commodity framegrabbers and video4linux. Even if you don't use Linux studying the OSS sites will get you the information you need. Hint, calibrate and count pixels.