Interesting developments in traffic monitoring

About a couple of years ago I had a conversation with someone who did a bunch of contract work on traffic monitoring systems - he was networking all the traffic cameras and radar together and providing a nice web interface with Traffic 511 like monitoring of congestion data. The added bonus was you could actually get a live feed of traffic on your favorite streets - his system was for Alameda county and hence monitored mostly city streets, not freeways. During the meeting he talked about how much cheaper his implementation was than the tens of millions CalTrans spent on their system which used data read from FasTrak transponders in people's cars. When he mentioned the trip time data that Traffic 511 was providing (and is now shown on electronic signs by the freeway) I asked him if they were doing the dumb thing of quoting actual measured journey times - it turns out that was how they do it.

I say it is dumb because if you are in Oakland (say) and trying to get to San Francisco and there is a wreck then you wont know about it from estimated journey times until someone has completed the entire trip - which could be a long time. And as each person arrives at the destination their data is as old as their complete journey time. Now a way around this is to break journeys into much smaller segments and integrate along the path continuously but apparently they weren't doing that either, except perhaps at very gross city to city levels.

Ideally as soon as there is a change in traffic flow past a point, as measured by velocity and number of cars at that point you should be able to feed that information into a model, combined with everything else you know about traffic on the roads and generate a new estimate very quickly - without waiting for cars to drive an entire segment. As a physicist by training that just seems like basic fluid dynamics to me... And you can feed in other information about known changes to the configuration of "pipes", ie. the roads. So when CHP calls in that they will shut down a freeway for 15 minutes to bring a medivac, or take it down to 2 lanes while they sweep debris, or open up Hwy 580 to trucks for a while then you should be able to feed that data in an make travel time prediction adjustments immediately. Thus the effect of a wreck in Oakland can instantaneously be feed back to someone in Milpitas traveling to Berkeley and allow them to choose an alternate route.

Anyway two years ago most of the clever stuff you could do really depended on having a lot of data about where cars were which either meant outfitting a lot of cars with telemetry systems or getting access to the FasTrak transponder data. Well of course two years on we have an increasing number of gadgets flying up and down freeways that already have GPS and data links built-in - they are called phones. So I kick myself for not thinking about what Nokia is doing in a trial where they are using standard N95 GPS equipped phones to see if they can predict traffic flow information - not just travel times, but predict information like I suggested. Plus the benefit is this system can be extremely cheap to set up since people like participating in such "social" applications (cf. Seti@Home and Folding@Home), even more so if they are actually of benefit to themselves. I think it will only be a matter of months before some open source project, quite likely sponsored by Google (and running on Android) will produce an app and centralized traffic monitoring/prediction services to the masses.

The only issue is that do really great prediction actually requires more than simply measurement - for this reason I think the initial data provided will be just the same level as Traffic 511, just measuring speeds on the freeway. To do really smart stuff requires complicated routing calculations and someone like Google (say) with access to that, and smart people to crunch the numbers and build models will be needed to build the software. All good stuff to build into Google Maps right and make it even more of a killer app - especially when mobile.