The Limit: Whether Or Not You're Detected

In your travels, you have no doubt seen something like this:

Such speed limit detectors usually show your speed in amber lights, unless it is over the limit, in which case the number flashes in red.

Have you ever noticed that these devices sometimes register your speed when you are riding a bicycle--but sometimes they don't?

Those of you who are engineers (or simply more tech-savvy than I am) may be able to provide an explanation of why.  I have always surmised that it had something to do with the position of the speed detector:  The ones placed above the traffic lanes don't register bikes riding on the shoulder or even the far right side of the traffic lane.   Then again, I've seen detectors on the side of the street or road--and, in a couple of cases, right in the middle of a bike lane!--that did not track me.

An article in the IEEE Spectrum--the journal of the Institute of Electrical and Electronics Engineers--may shed some light on why this is so.  

In it, Spectrum contributing editor Peter Fairley describes the advances made in detection systems used in robotic (self-driving) cars.  The best such systems could spot only 70 percent of cars just a few years ago.  Now that figure is nearly 90 percent.  Similar improvements in noticing pedestrians, and even birds and squirrels, have come.

In contrast, those same systems can detect a bicycle from 59 to 74 percent of the time.  While that is an improvement from a few years ago, the advancement doesn't come close to what has been accomplished in tracking other beings and objects.  

According to University of California-Berkeley research engineer Steven Shladover, "Bicycles are probably the most difficult detection problems autonomous vehicle systems face".  Down the coast, in UC's San Diego campus, visual computing expert Nuno Vasconcelos offers a possible explanation as to why:  "A car is basically a big block of stuff.  A bicycle has much less mass" and, he says, "there can be much more variation in appearance--there are more shapes and colors and people hang stuff on them."

Vasconcelos' explanation makes sense, as far as it goes.  In the article, however, Fairley posits that part of the problem is how those systems are "trained".  Actually, they "train themselves", if you will, by studying thousands of images in which known objects are labeled.  Most of the training, Fairley says, has concentrated on cars, with very little on bicycles.  

I will not address the question of whether robotic technology can or will replace human drivers:  Folks like the ones Fairley cites (or, for that matter, Fairley himself) can say far more about it than I can.  From what I've read, however, I believe I can surmise that self-driving technologies have a better chance of working on Interstates, Autobahns, Autoroutes or other high-speed, multi-lane highways where bicycles are seldom seen, or are banned outright. 

Likewise, detection technologies have a better chance of detecting speeders on such highways than they have of catching me riding my bike over the speed limit on city streets!