Smartphone-Based Real-Time Travel Mode Detection for Intelligent Transportation Systems

Abstract

Intelligent Transportation Systems include all transportation modes, aiming to improve the efficiency of transportation in many situations. Identifying the transportation mode of users is a key performance and quality requirement for ITS. Many works have proposed the extraction of this contextual information from large datasets, in an offline manner, while fewer works have proposed techniques for detecting this type of context in real-time. Real-time detection could allow cost and latency reduction for ITS applications since all the processing can be made within smartphone devices and actions can be taken quicker, as the information will always be up to date. In this paper, we propose a real-time travel mode detection technique that applies supervised machine learning (ML) on location data extracted from smartphone sensors. We evaluate its performance through confusion matrix metrics obtained from inferences made during field tests with 37 users in the metropolitan area of Rio de Janeiro, Brazil. A total of 2519 samples were classified using a prototype application, namely CityTracks-RT, in which we implemented the proposed technique, using only smartphone hardware resources. Additionally, we evaluate the performance of a popular off-the-shelf activity recognition (AR) solution in the travel mode detection task, to justify the development of specialized techniques. For this, we implemented another prototype application named CityTracks-AWARE. Thus, it was possible to verify that generic AR solutions are not sufficient for travel mode detection, through the performance evaluations carried out, and that our proposed technique provides a good basis for a specialized solution.