“Memo #167445/A — To the T-800 Production Lines. Guys, love the new designs and features, but the overall BOM is still too high. Get in contact with supply chain management and cut the overall bill. — Skynet 14.06.2026”
While I was attending the recent Embedded Vision Summit, I heard a talk by Google about their new neural network technology. With their trained neural networks, they can now identify objects and people with higher accuracy than can than human beings. As part of this presentation, they showed some lovely examples of things like small children cuddling teddy bears, plush fluffy poodles, and apple trees in flower.
At the end of the presentation, the floor was opened to questions. One query that caused some nervous laughter was when Embedded.com's Editorial Director, Max Maxfield, noted that when the first Terminator movie came out in 1984, the idea of an artificial intelligence (AI) like Skynet attempting to take over the world was considered by most to reside purely in the realm of science fiction. Max then asked if the presenter ever woke up in the middle of the night envisioning the possibility of a time 20 years hence in which we are all cowering in our basements listening to our robot overlords stomp past.
Some time later, I found myself chatting with Max about Basler 's digital cameras. (As an aside, I started off as a particle physicist at two of the world's great particle colliders. Following a short career as a software developer, I moved over to new business development at Basler AG — a leading global manufacturer of digital cameras for industrial and retail applications, medical devices, and traffic systems. I am also a big fan of science fiction films.) During our conversation, I happened to mention that you can have the best camera in the world, but that many users mess things up by employing cheap, low-quality cables to connect everything together. I concluded by saying: “The way to defeat Skynet is to connect everything together with cheap cables.” Max laughed and said that I should write this up as a column for Embedded.com, which is how we come to find ourselves here. Consider the following scenario…
An alternative timeline: May 2016. The T-800 ID#9420 Terminator's eye camera has just absorbed an image of John Connor through its liquid lens. John is rushing towards it not looking particularly happy (exact emotional classification pending). The lighting fast CMOSIS 4MP sensor is shoveling 90 fps through the buffer-less FPGA running sophisticated image-processing algorithms that dramatically improve the picture quality.
Meanwhile, T-800's brain is running the latest release of its neural network software, which has been extensively trained on hundreds of thousands of pictures of angry and scared human beings, especially John. The only thing preventing this perfect killer app vision system from being able to decide if John is wielding a pickaxe or a large baguette is “A great deal Skynet found on the Internet.” That is, the cheap-and-cheerful cables linking the T-800's eye and brain subsystems are having much the same effect as does Heathrow Airport's passport control when you are late connecting between flights.
Obviously, the T-800's data transfer protocols have error correction, but this is not good for your frames per second (physics is a bugger for this sort of thing). Without shielding, electrical noise from the T-800's other subsystems may decimate the signal transfers, in which case the error checking will just keep asking for a resend until a good image eventually limps its way through the system.
Clearly the designer of any system doesn’t want overkill (although maybe this is not a good choice of words in Skynet's case). If your application is tasked with monitoring sunbathing sloths on a slow Sunday afternoon, you may be able to get away with saving a bit of cash on your cables. However, if the quality of your cables is the only thing standing between you and a stern reprogramming with a pickaxe… then maybe not.
Now don't get me wrong — I'm not against this technology. At Basler we believe machine vision based on deep neural networks will offer tremendous benefits in many different roles. I'm personally using the excellent color quality offered by our PGI feature to measure skin pigmentation. In case of a zombie apocalypse, I'll be wiring up a sentry gun to a camera coupled with a killer zombie app, and you can bet that I'll be choosing the best cable I can find. (Note to self: When identifying Zombies, remember that they're typically a bit greener than the average human being.)
Mark Hebble prior to the zombie and/or robot apocalypse (Source: Mark Hebbel)