I will have one more work week in Texas after today. I enjoy my job and the people where I work a lot and it was agonizing to turn in my notice. Part of the job I love the most is the requirement to create sophisticated machine vision and video analytics applications with cheap USB cameras and ARM embedded computers that run embedded Linux, usually Debian. We prototype a lot of the stuff on Raspberry Pi’s which is great because there is such a big user community it is easy to quickly get answers about just about anything. There are four cameras in the article accompanying this post that range in value between $20 and $50.
All of the cameras work just fine right out of the box for the purpose for which they were design–that is generally streaming video with camera controlling the capture gain and offset. Conversely, it reduces the repeatability and precision of most machine vision application if the offset, gain and lighting controls are not managed by the application. So, it has been part of my job to dive into the driver code far enough to figure out how to set the registers that need to be set to control cheap cameras well enough to work in accord with the stringent requirements of many machine vision applications. That takes a lot of patience and, although it is not exactly rocket science, it is very rewarding when the last piece of minutiae is chased down and the stuffs starts working.
One thing I have learned is that this “big data” thing is here to stay, at least in my world of machine vision, embedded computing and video analytics. There are tons of things you can almost do deterministically that become tractable when enough data and machine learning are thrown at them. I am loving working with Weka and R and the machine learning functionality in the OpenCV library because they open up new vistas, not to mention I can more frequently say, “I think I can do that” and not squint my eyes and wonder whether I am lying.