 Parking is frustrating. I probably speak for most people in pointing out that this is not an activity that I enjoy. It wastes time, it wastes money in the form of gas, it's polluting and irritating, and at the University of Buffalo we have a parking problem. I suspect many universities do as well. Ours is somewhat exacerbated by the design of the campus, where we have a long, thin academic spine, which is where everybody's trying to go. That spine is surrounded by an increasingly large ring of parking lots, but finding a spot is something that's very challenging, wastes a lot of time for students and faculty members, particularly around the time the classes start and end. We approached this problem. There have been many, many attempts to work on parking to instrument lots in various ways, and this is not a problem I probably would have worked on, except for the fact that my student, Anand Nandagudi, former PhD student of mine, came to me and said, you know, I think we can approach this problem in a different way. And his idea, which I think is very clever, was to do it in a way that requires no additional infrastructure and no human input. The existing approaches to this problem, I mean, you know, you can throw infrastructure at it. So I can instrument the entrances of lots, using sensors or gates. That requires that I have to instrument every part of the lot, every entrance or exit to the lot, so that gets expensive very quickly, actually. We've looked at commercial solutions to this problem, and they are not cheap. I can also throw other forms of infrastructure at it. So I could put cameras on every light pole in the parking lot. Maybe some parking lots have this already for security reasons. Ours don't, so this would be another case where this would be a potential, you know, additional extra infrastructure I'd have to throw at the problem in order to solve it. The other thing I could do is I could try to get people to participate in the process of monitoring these lots. So when you arrive at the lot and park, you could get out your smartphone and say, hey, I parked, and then I could record that information, and when you left, you could do the same thing. But humans are unreliable, and we didn't want to rely on people to do that. So infrastructure is too expensive. Humans are too unreliable and we don't want to bother them. How do we solve this problem with no additional infrastructure and no human inputs? The core observation here is that there are interesting transitions that we can exploit. So when I'm in my car, it's my attempt to drive in a car, and then I get out and I start walking, if this transition happens inside a parking lot, then I assume that this person has parked. They drove into the lot, got out, walked away from their car. This is a parking event. And so there now is an additional car in this lot. On the other hand, when somebody goes from the other direction from walking to driving, then I assume this is sort of de-parking or they left. They left a spot. And now there is an additional spot in this lot. And this was the core observation that allowed us to build this system because it turns out that the accelerometers on modern smartphones are capable of detecting these types of activities and therefore are capable of detecting transitions between those activities. So Pocket Parker starts with leaving the accelerometer running and doing activity recognition. This is not something that we had to build. We built an earlier prototype, but eventually we just moved over to using the Google Play Services Library, which does this very accurately automatically. So we monitor whether the user is driving or walking. When they go from driving to walking, we assume that they parked. Inside a lot, we assume that they parked. If they go from walking to driving, inside a lot, we assume that they left a spot. And this is the core sort of component. Now, we do need to figure out where they are. And the nice thing about this system too is it allows us to trigger use of the GPS so that we don't have to leave it running all the time and consume a great deal of the user's battery. So when I see one of these transitions from driving to walking, I flip on the GPS and I grab a geolocated point that indicates where that transition took place. I use mapping databases to figure out where the parking lots are. And then if that transition occurred inside a lot that I'm monitoring, I adjust the lot capacity accordingly. And so this is a nice system in that, again, it requires no user input. The user can leave the phone in the pocket or in their bag. We tested multiple different locations for the phone and found that this transition was pretty accurate, regardless whether the phone was in a pocket or a purse or somewhere else. And there's no additional infrastructure required. I can deploy this system without, you know, spending a dollar on gates or cameras or sensors or any of the sort of expensive system. So this was, you know, the core observation and this was the beginnings of a system that we called Pocket Parker. We also consider this to be an example of something called pocket sourcing. Crowd sourcing is this beautiful idea where I can collect information from lots of individuals and use it to find out things about the world. In pocket sourcing, we try to do a little bit better in that we don't rely on manual human inputs or really any user interaction at all. The goal is how much can I find out about the world using a smartphone that is safely ensconced in somebody's pocket.