Lost & Foundby Omer Shapira / NYU
The Science of Getting Lost in a City, and How It Can Benefit Future Technology
This project's goal is to study the science of getting lost in a city setting. When is someone 'lost' on the street? How does a lost person look when plotted on the map? Is there more than one kind of 'lost' behavior on a map - and if so, is it different between - say, lost alzheimer's patients, lost children and lost tourists? I suppose there is, and I suppose that it can be found.
Human dynamics in cities are radically different than movement in an open terrain. The line of sight is trim and constantly changing, new dangers are popping up by the second, and the perception of space is severely contaminated by sound and smell interference.
This project will try to find an answer to the these questions:
- Can a computer indicate that a person is lost, based on his movement in the past n minutes?
- Can a human's motion pattern indicate that he might have Alzheimer's, or some other memory/cognitive condition?
- Can city planners learn to avoid mass confusion, using current statistics of poor navigation in a space?
Using the data (and ultimately, the knowledge) acquired in this set of experiments, we will be able to help many different groups: Children/elderly people (and their families), city planners who want to know what parts of town don't work, Psychologists, Computer Scientists, and people who are asymptotically stuck on the street, wondering what to have for dinner.
Commute data will be scanned by unsupervised learning algorithms to detect irregularities in a path. If we have enough of the same route, that route will be added to the research. Irregularities that appear over time will be studied to be detected in real time.
Data security: The research will begin in New York University's computers. All of it will be done on secure servers, inside secure storage databases.
Privacy: All data is anonymized. Only common route data is used, therefore the route from a residential building to the subway is highly unlikely to end up in the research. We're only looking for the most frequent A to Bs.
All of the findings, from machine-learning output to proper academic babble - will be available online on the project's website. Data will be categorized by geographic region, so city planners can access areas of their interest.