I answered a question on the APBP listserv and thought I should capture it and update it as we learn more about the technology.
The XX State Highway Administration is telling my hometown that updates to signal hardware, namely, video detection technology, will be able to detect bicyclists where loop detectors are installed currently, and not doing a great job. I’ve noticed in practice that it seems to work, but I wonder about night time and positioning sensitivity, etc. What’s the experience out there with video detection for bikes?
Also, they rejected out of hand the use of a bike push button actuator installed at the right curb edge as not permitted based on the XX version of the MUTCD. I thought I recall Portland and other jurisdictions had installed bike actuators on the curb.
Well, since he called Portland out and I am that guy in Portland, I felt obliged to spend the lunch hour pounding this out.....
Video detection is better than you might think and I have never heard of a ban on push buttons, but I wouldn't encourage their use either because of the potential for them to get hit if they are close to the curb. We sent an intern out to answer your question a few years ago and the informal, limited sample size, nonscientific results are posted here. The bottom line for traffic design is we can and should do better at traffic signals. For those on the listserv that want details, keep reading...
The City of Portland has studied several different types of detectors at a few select locations. Often, we have a unique problem we're trying to solve.
Our standard for detecting people on bicycles is the same as it is for motor vehicles, the gold standard of inductive loop detection. We have used this for a long time and the technology is proven. The use of inductive loops requires the agency ask the field technicians to set the sensitivities high enough to identify bikes. Yet, it does require people on bikes to be over the loops and they are problematic for detecting people on bikes (as a designer) because you have to know where a person on a bicycle will be. We use stencils on the pavement to help give an indication of where a bicycle should be placed to be detected, but this doesn't work so well in snow, when leafs are on the pavement, etc. Bottom line, all detection has some challenges and inductive loops are no exception. One thing we have tried to improve is to provide positive feedback to someone that they have been detected. To accomplish this, we have implemented a detector confirmation light to give an indication that they have been detected. Some information about that is found at the following link.
The City hasn't used video traditionally because of expense and problems with fog, false calls, etc that were originally documented in this research by Purdue University . That hasn't stopped many agencies from widespread adoption of video as their standard, but it did lead to the Indiana DOT eliminating the use of video detection. Now, this research is from 2006 and the technology has improved a lot, but we still use inductive loops in most cases in Portland. Let me say it again, video detection has improved. Our recent study (not really research) showed that video detection worked better than we anticipated (nearly 100%) with good lighting conditions at the intersection and no fog. We watched some video during different times of day including periods of darkness where we thought we would see significant problems and it was better than we anticipated for detecting people on bicycles. This study did a one to one comparison with microwave detection, which we thought would be better because of the properties associated with it as a technology and it was similar in terms of performance. The link has some numbers, but it is not a definitive study. The National Cooperative Highway Research Program 7-19 is going to study this further.
A nearby agency that uses video detection for most of their applications has decided to stop using it because of the increased maintenance associated with problems reported by people about the detection failing in the on condition (basically side streets getting green time when there is no traffic), which is one of the problems that Purdue University found in their original research.
Our next step is to study a thermal imaging camera because we are interested in the reduced maintenance potential associated with not having to clean the lens and the ability to distinguish between cars and people on bikes. We just installed it a few weeks ago and have some initial concerns, but nothing yet in terms of a study to share.
We also have a location where we are going to put the new Sensys bike detection product to use because that system is wireless from the in ground unit to the traffic signal controller, so it may reduce our costs. We're not sure how well that will work, but we're going to study it.
To answer your original question, the position sensitivity is a big part of making any detection work. Video and microwave are both more flexible in terms of the field implementation obviously. Our technicians care about people on bicycles and they have worked to make the technology work, it is not something that is particularly easy to setup.
Sorry for the long response, this is clearly an area where more research is necessary.