Machine Vision for Robotic Applications

A few weeks ago I was presented with an opportunity to work with Donnie Franzen on a project that required a variety of disciplines; robotics, optics, computer networking, etc. Although I won't be sharing intimate details of the project, yet, what I will share is my research thus far.

The system that Donnie has proposed is one that requires a digital camera to follow an object in motion, to be able to record video, potentially stream the video to the web, and be operated remotely (oh, it has to be dummy proof too). Our initial idea is to build a kiosk that an end-user can operate from a high level. The front-end, C# or Flex, will control the transmission of the camera ad the system as a whole. I know a bit about image processing and video capture in C, it can be pretty messy. When I stumbled across The Imaging Source, in pursuit of a camera (DFK 21AU04), I was delighted to find a nicely abstracted suite of APIs for image capture. Download the trial version and you'll find a handful of projects in VB6, VB7, C#, and VC. Although you don't have to use Imaging Source hardware, a license for their software comes with the purchase of a camera. Why not buy a camera, get the license, then add additional cameras of your own. But I must say, this camera seems to be super beefy and sturdy, we'll likely stick with Imaging Source hardware.

There are a host of projects included in the SDK. Everything from basic image processing to video recording. I haven't spent a ton of time digging in but like I mentioned above, The Imaging Source did a fantastic job abstracting that which is unnecessary and only exposing specific elements of the video object.

Here is a sample of the video quality recorded uncompressed at 640x480.