Poster: Enhanced collaborative sensing scheme for user activity recognition

Accelerometer data on a user's mobile phone contains abundant information that can be employed for user activity recognition. However, the existing schemes cannot provide accurate inference results under moving environments (e.g. on a train). This is because raw sensor data changes with the motion of the environment as well as the user. In this paper, we propose an enhanced collaborative recognition scheme that exploits neighborhood sensor data shared over a wireless network. Our preliminary experiment showed that the acceleration data of two different subjects on the same train was highly correlated. Further, it was possible to detect whether two subjects were on the same train with an accuracy of 69%. This result indicates that raw sensor data shared by spatially neighboring users can be used to detect transportation modes and enrich social networking applications.