Processing mobility traces for activity recognition in smart cities

Abstract

Human mobility modelling has emerged as an important research area over the past years. The opportunities that mobility modelling offers are widespread. From smart transportation services to reliable recommendations systems, all require generation of mobility models. Since mobility of humans is generally motivated by the activities they perform, activity recognition emerges as a vital initial step towards building better and accurate mobility models. The activity recognition can be carried out by analyzing relevant data from GPS devices, accelerometers and many other sensing sources. The most common approach is to combine data from different sources, analyze that data and recognize the type of activity being performed. However, this requires access to many specialized devices and customized infrastructures. As an alternate, this paper introduces a novel approach to recognize activities from the GPS traces only. This approach utilizes Adaptive-Neuro-Fuzzy Inference System (ANFIS) which combines the power of neural networks and fuzzy logic to recognize activities. The approach is tested on three different datasets and shows promising results. In addition to this a multi-cloud architecture is proposed, for the deployment of such a system.