Today was the first day of the Mobiquitous 2014 Conference. Like at many conferences this was Workshop Day, and I attended the i-Locate Workshop.
The workshop was pretty nice. The first talk was right into the topic of indoor localisation. The talk was titled “Selective Mixture of Gaussian Clustering for Location Fingerprinting” by Nguyen and Luo. In the talk he basically showed that cell localisation using clustering with Gaussian Mixtures performs better than k-means and fuzzy k-means. So, I would expect that result to some degree, and also there was no comparison to SVMs inside. Also like all these methods they have the classical problem of finding a good k. However,the talk was good and the discussion also really good, and showed the good experience of Khuong (Nguyen) in that area. We discussed problems of building up the radio map (they do it minimally once a month, or if many things moved), sampling resolution (they sensed every 15cm, with 8 different position for a person shadowing the signal) and others.
The second talk was “Mobile Indoor Localisation in Museum Environment: An Experimentation in FIBAC Project” by Acanfora et.al.. It build on the results of the first talk and also compared cell localisation to triangulation. You could see how raw triangulation on WiFi still is considerably bad. They did however build nice additional models like a path recommender, and modelled typical movement behaviour if I understood that right. Also the discussion was nice.
Both projects based their work on WiFi data/fingerprints btw., and both tested the “harder” case, that is movement along the walls in a crowded environment.
Especially in the case of triangulation, that is you want to compute a point in space, however, I would have found it cool if they would have added models which incorporate movement like f.e. particle filters or similar.
After the coffee break the next talk was “Personalised multi-modal route planning: a preference-measurement and learning-based approach”. In a quite large scale (1500 ppl) survey they assessed preferences for persons deciding on witch mode of transportation to take (that is f.e. travel time vs. costs and so on) to build a standard model of preferences for the average user of Denmark. Building on that model a Bayesian network was created to predict a users choice based on his preferences. A lot of that interesting talk was about the update of the believes of the internal preferences, as this is a computationally hard task (especially in that setting).
The main problem with that approach is that you have to come up with categories the user has to have preferences on and correctly update them. To overcome that I wonder if collaborative filtering can be used (I suggested that, and want to try that out by myself too).
The next two talks had privacy in there mind. They were “Protecting data resources and users in GeoData for interior and exterior spaces” by Cadzow and “Applying Archeological and Forensic Science methods and experience to outdoor and indoor mapping” by Cadzow and Cadzow.
As some of you know usually I find privacy: Very Important and none of my business as this is its own complex stream of knowledge I do not follow (ok, I know how to overwrite the stack in C programs and so on, but that only leads me to admire the profession). In this talks however a new nice viewpoint on privacy of locations was raised. Mainly how can you bring thinks like a “No Access” sign on a door into the digital world, and how much detail should you be able to query at a distance. Both relevant and interesting questions. The archeological talk also introduced interesting methods to survey an area, that is the magnetometer survey. I wonder how that could be used.
After the good lunch the last talk really introduced the i-Locate project “i-Locate: Indoor/outdoor Location and Asset management trough open GeoData; rationale, scope and objectives” by Conti. This was about the recently started EU Project i-Locate which is basically about doing something like OSM for indoor. I like the idea, but, as I think at least some people in the consortium agreed beside standarts there must be a working toolset and best practices as Cadzow put it for us developers and researches to use to really have an impact. It will be a project to monitor!
Beside the talks the audience was awesome and I met many interesting people beside the speakers! Quite some went home already and did not stay for the main conference, I still hope I can keep contact with them.
So, that was my wrap up for day 1. Let’s see how day two will be!