I haven’t blogged in a long time, so I’m looking for any excuse. Coming across a really cool paper seems a good one.
The paper is from Romit Roy Choudhury’s group at Duke, and appears to be a collaboration with some researchers at EJUST in Egypt. The title is No Need to War-Drive: Unsupervised Indoor Localization, and was published in MobiSys (one of that groups’ 4 papers), which is taking place right now.
The main idea is very clever. Basically, dead reckoning on a mobile phone sucks, but there are lots of ‘landmarks’ that can be established from various sensors (e.g., magnetometer, sound, ambient light, etc.), that can be used to ground the estimate. These landmarks are discovered automatically using clustering. If the goal is to place a user on a map, then they can also use certain preexisting landmarks. For instance, they are able to accurately detect when a user moves up or down an elevator, escalator, or staircase, and so they can use the known locations of these structures to anchor landmarks to a map. They did some thorough testing in two buildings on campus as well as in a mall, and the accuracy is exceptional. The paper is written somewhat informally, but tells a great story of how the system was developed. There isn’t a strong novel contribution, but a fantastic combination of some clever tricks to build an indoor localization system that requires no calibration.
Kudos to the authors on a great paper. For anyone thinking about indoor localization, I recommend this work.