Online recognition of people's activities from raw GPS data

Abstract

The widespread use of GPS devices leads to an increasing availability of people traces. The GPS trajectory of a moving object is a time stamped sequence of latitude and longitude coordinates. The analysis and extraction of knowledge from GPS trajectories is important for several applications domains, ranging from traffic management to advertisement and social studies. We present an approach capable of incrementally extracting semantic locations from people's trajectories and inferring the activity done by users. We associate the places visited by people during their movements to a meaningful human activities using a novel algorithm that cluster incrementally user's moves into different type of activities. Studies using GPS records from a confined spatio-temporal region demonstrate that the proposal is effective and is capable of inferring human activities without depleting the phone resources.

Online Prediction of People’s Next Point-of-Interest: Concept Drift Support

“…Unfortunately, these works do not permit incremental training of the models, since it is based on a posteriori learning, secondly, the fact that authors propose matching functions in form of strategies (simple, polynomial, logarithmic and aggregation strategies) can create a computational complexities; difficulties to choose and set the rights parameters for the right strategy. Sébastien et al extended a previously proposed mobility model called the Mobility Markov Chain (n-MMC) 2 , in order to keep track of the n previous locations visited [7]. This proposal essentially corresponds to a higher order Markov model.…”

“…Assuming that we track a user outdoor activities using an incremental solution such as that one presented in [2], every activity (location) detected is called a Place of Interest (POI). A POI is an urban geo-referenced object where a person may carry out a specific activity.…”

“…This dataset recorded a broad range of users' outdoor movements, including not only life routines like go home and go to work but also some entertainments and sports activities, such as shopping, sightseeing, dining, hiking, and cycling. We treated the dataset to add some information to the database like user's speed and orientation to be used in our previous work [2] to detect users' activities from their GPS traces, every POI recognized was identified by its unique combination (longitude latitude), the result was organized in dataset (see table 1) as user's daily routines. …”

“…We tested our algorithm using an Android smartphone from Sony (Sony Xperia S) with 1 GB of Ram and 1.5 GHz dual core processor. The first test concerns the RAM usage, we added our solution to our precedent work [2] where we recognized incrementally users' activities, then we compared the set with a well-known GIS solutions "Waze Social GPS Maps & Traffic", one of the best free navigation applications that won the best overall mobile app award at the 2013 Mobile World Congress, the reason of such selection is that Waze has a lot in common with our approach. In fact it gathers complementary map data and traffic information from its users like police traps (can be seen as a POI in our case), and learns from users' driving times to provide routing and real-time traffic updates.…”

“…One of these important research topics is the prediction of user's next location [5,7,16], where results of these research are used in a wide range of fields like traffic management, public transportation, assistance of people with special needs, commercials and advertising. Our precedent work [2] proposed an online activity recognition system that offers the possibility to understand what people are doing at a specific moment by inferring incrementally people's interesting places from raw GPS data. In this paper, we are proposing an evolution of our precedent system that estimates the users' actions in the future by predicting their next visited point of interest.…”