Geographic Information Systems Stack Exchange is a question and answer site for cartographers, geographers and GIS professionals. It only takes a minute to sign up.
Sign up to join this community
I'm looking for a benchmark GPS dataset, available for research purposes for free. I've found GeoLife GPS Trajectories dataset from Microsoft Research but I find it a little incomplete.
What I need is the GPS activity data of a person, as in (latitude, longitude, date) tuples, tracked for at least several months, preferably continuously. I would also like the recordings to be non-sparse; at most 1 minute between each record.
Can you point me towards such a reliable dataset?
I think your best chance will be to track yourself. If the idea bothers you, that's the reason why you won't find such data public anywhere.
The best I can think of is the GPS traces available from OpenStreetMap. They are not going to be continuous, but there are a large number of them.
On the OSM website select "GPS Traces" to have a look at what is available for a particular area.
One choice is to draft up a contract and hire a lot of people. Provide them with GPS units configured to take readings providing the data you need, enough batteries to last the contract, and instructions (plug it in with this cable to upload nightly, email me this file, etc.)
You'd definitely need to write in the contract how you would restrict the distribution of the data and anonymize it to protect it (perhaps providing a rough half-mile radius of exclusion around the points the person indicates are private,) and you might even consider buying insurance against loss. If traces of peoples' activity became public, they would be filled with information such as, "I leave for work every morning at 7:00 and come home every night at 19:00", and a plot would look like a giant asterisk centered on their house saying "rob this place between 8:00 and 18:00." You can see why you'd need to be concerned about privacy and security.
If you think about it, you're asking for some very expensive data. And without a statistically large enough set, it's going to be of dubious value. Think of how different traces would be between a construction worker (a new repetitive commute after every completed building), a postal carrier (a very repetitive and very serpentine route), an office worker (a mostly repetitive direct route), and a tow truck driver (new routes continually.) Socioeconomic status might impact the traces: lower incomes might follow public transit lines more and travel less. Parents of school-aged children might have average higher after-work commuting miles. Not to mention the guy who drives the Google Street View cars.
None of those traces are likely to intersect any of the others in any meaningful way.
The number of unique styles are likely to be finite, but so high as to require a significant budget to obtain. And that would be in just one city.
You might be able to obtain a smaller (cheaper) set of data if you defined your goals better. If you're trying to quantify the various types of patterns, maybe you sample a broad range of people in a variety of cities. If you're trying to figure out who would benefit from mass transit, or where to lay commuter rail corridors, you're probably better off counting cars on the various roadways around the area you're planning to serve and conducting surveys.
I wouldn't hold my breath. Data at such precision would be a massive undertaking and have enormous privacy implications (even if only for 30 days for one individual that would include 43,200 data points (if recorded every minute), and would undoubtedly identify there home location).
If your interested in substantive questions that such data would hold this advice won't help. But if your only interested in some type of analytical strategy to handle such massive data you should be able to simply simulate data at that scale to serve whatever your purposes are. To simulate data I would suggest you take a look at the R statistical program, and the spatstat and the trip package in particular (as well as all the spatial modules in R).
I would be skeptical even animal tracking data would meet your requirements for data points in such short intervals. I could list a few articles I have read that use cell phone data to estimate human activity patterns, but none I have read would come anywhere near that long in time or measuring individuals activity that frequently.
Open PFLOW project offers:
open dataset for typical people mass movement in urban areas
Tokyo metropolitan area is available and Chukyo metropolitan area seems to be under preparation.
Details can be found in a recent publication:
Takehiro Kashiyama, Yanbo Pang, Yoshihide Sekimoto, Open PFLOW: Creation and evaluation of an open dataset for typical people mass movement in urban areas, Transportation Research Part C: Emerging Technologies (2017) Volume 85, Pages 249–267.
T-Drive trajectory dataset is a recent find. It provides:
a one-week trajectories of 10,357 taxis. The total number of points in this dataset is about 15 million and the total distance of the trajectories reaches 9 million kilometres.
Although not about human movements, Liquid Robotics company makes available interesting dataset from its PacX challenge. Data about location and environmental sensor readings of four robot gliders sailing through Pacific Ocean are available for download. More info about (really cool) project on the blog, via WIRED and this talk.
Another option to tackle the privacy issues would be to use animal tracking data. I guess data protection will be less of an issue here. As an advantage, you might still be able to test your software/methods with real world movement data. Disadvantage might be that if your application needs 'human specific' movements - they might not fit your purpose.
Have a look at Movebank or DRYAD websites to check if some of their data might fit into your project.
Yet another option is the spatial part of the Chris Whong's NYC taxi data. They only provide pick up and drop off locations, however the volume (11 GB!) and contextual info (fare, passengers, etc.) make them really attractive (alternative download, more info about privacy concerns raised by the data).
Urška Demšar's post on her recent paper on 'Analysis of Human Mobility from Volunteered Movement Data and Contextual Information' promises:
There will also be a free data set of volunteered GPS trajectories linked to this paper available soon. Stay tuned.
Update: paper mentions that data will be available on CRAWDAD mentioned by @ejel but I havent found it there.
Another option might be to create synthetic dataset yourself. If you need some inspiration look at recent paper by van Dijk J (2018) Identifying activity-travel points from GPS-data with multiple moving windows Computers, Environment and Urban Systems (link). More details are provided in paper's appendix and code and example dataset are available on github.
Tahina Expedition (Google Earth Blog) http://www.tahinaexpedition.com/map has being sailing around for most of last year now.
KML can be processed http://maps.google.com/maps/ms?source=embed&hl=en&geocode=&ie=UTF8&t=k&msa=0&output=nl&msid=103005318482134016767.0004670ab348ba9fa7b1f [was a gps track now converted to kml]
I'm also looking for the exact type of dataset you're looking for. Unfortunately, so far I haven't found one yet. Despite GeoLife data, another source I found is CRAWDAD. The site has a GPS logs from San Francisco cabs and also New York pedestrians. Unfortunately, for NYC pedestrians they only provides relative coordinates rather than lat/lon.
There are many research topics where the data necessary to answer the question are unavailable for moral reasons, and experiments which overstep these bounds may lead to future restrictions, as was the case with the Milgram experiment. More recently, AOL had to pull a corpus of search queries because of the privacy concerns, and the only reliable dataset we have on email habits came from the Enron trial.
So while its entirely technically possible to get such a trajectory dataset, it may never be practical due to the privacy implications. As other answers have mentioned, relative datasets, aggregation over individuals, or simulation may all be better approaches to address your question, while avoiding the privacy issue.
Recent news shows that the IPhone creates a long running track log. Perhaps you could find particpants who would be willing to let you use the data.
People provide that data to Google for free around the clock. It's called Latitude. Maybe they will share it as generously as their users have shared it with them.
