Accelerometer Based Transportation Mode Recognition on Mobile Phones

Abstract

Recognizing the transportation modes of people's daily living is an important research issue in the pervasive computing. Prior research in this field mainly uses Global Positioning System (GPS), Global System for Mobile Communications (GSM) or their combination with accelerometer to recognize transportation modes, such as walking, driving, etc. In this paper, we will introduce transportation mode recognition on mobile phones only using embedded accelerometer. In order to deal with uncertainty of position and orientation of mobile phone, acceleration synthesization based method and acceleration decomposition based method are introduced. Performance comparison indicates that acceleration synthesization based method outperforms acceleration decomposition based method. We will discuss the factors affect the recognition accuracy of acceleration decomposition based method and present potential improvements.

Modelling user situations with smartphones

“…As we want it to be automatically annotated, we should use the smartphone's sensors to infer the transportation model. Fortunately, many previous studies have shown promising results on inferring transportation modes [8] [9]. This study will borrow some ideas from the results of the previous studies.…”

Transportation mode detection – an in-depth review of applicability and reliability

“…A direct consequence of the difference in focus between disciplines is the types of devices that are used to collect data and / or perform the mode detection. While there is a prevalence for using smartphones in LBS (Manzoni et al, 2010;Reddy et al, 2010;Wang et al, 2010;Hemminki et al, 2013;Montini et al, 2014;Shah et al, 2014;Yu et al, 2014), TSc and HG mostly use dedicated devices such as the Geostats Geologger (Chung & Shalaby, 2005;Tsui & Shalaby, 2006) or other types of dedicated devices (Alvares et al, 2007;Palma et al, 2008;Stopher et al, 2008;Bohte & Maat, 2009;Schönfelder et al, 2002;Rocha et al, 2010;Zheng et al, 2010;Biljecki et al, 2013;Rasmussen et al, 2013Rasmussen et al, , 2015 with a few exceptions that use smartphones (Schüssler et al, 2011;Montini et al, 2014). This is an important distinction because, with a few exceptions (Stenneth et al, 2011;Shah et al, 2014), the studies that collect data using smartphones have a thick-client architecture, in which the client performs most inference operations and the server acts only as a storage backup, which contrasts those that use dedicated devices, which use a thin-client architecture, in which the server performs inference operations in addition to the data management.…”

“…While the data size available for each research group varies (150 hours from 16 users for Hemminki et al, 2013, 12 hours from 7 users for Wang et al, 2010 and an afternoon from 4 users for Yu et al, 2014), the precision and recall values are computed on accelerometer samples, which are labeled periods of a predefined duration where one accelerometer reading can be found either in one period only (tumbling windows) or in multiple adjacent periods (sliding windows) -(8 seconds for Wang et al, 2010, 1.2 seconds for Hemminki et al, 2013, and 10 seconds for Yu et al, 2014) associated to one activity. The validation of the proposed methods is done by allocating a percentage of the collected data set for testing purposes and training on the remaining set (Wang et al, 2010;Yu et al, 2014), or by using a leave one user out cross-validation (Hemminki et al, 2013). The advantage of using an accelerometer only approach to transportation mode detection lies in the promptness of response and battery efficiency of the method.…”

“…Research that focused explicitly on accelerometer only transportation mode detection report a precision of 70% for six transportation modes using decision tree classifiers (Wang, Chen, & Ma, 2010), 80.1% accuracy and 82.1% recall for seven transportation modes using AdaBoost together with Decision Tree classifiers (Hemminki et al, 2013) and 92.5% accuracy for five transportation modes using a Support Vector Machine classifier (Yu et al, 2014). While the data size available for each research group varies (150 hours from 16 users for Hemminki et al, 2013, 12 hours from 7 users for Wang et al, 2010 and an afternoon from 4 users for Yu et al, 2014), the precision and recall values are computed on accelerometer samples, which are labeled periods of a predefined duration where one accelerometer reading can be found either in one period only (tumbling windows) or in multiple adjacent periods (sliding windows) -(8 seconds for Wang et al, 2010, 1.2 seconds for Hemminki et al, 2013, and 10 seconds for Yu et al, 2014) associated to one activity. The validation of the proposed methods is done by allocating a percentage of the collected data set for testing purposes and training on the remaining set (Wang et al, 2010;Yu et al, 2014), or by using a leave one user out cross-validation (Hemminki et al, 2013).…”

A Versatile Annotated Dataset for Multimodal Locomotion Analytics with Mobile Devices

“…For example, Microsoft's GeoLife dataset [1] provides an impressive number of 17621 trajectories of 182 users over three years but the data is not annotated and only contains GPS traces. Wang et al [2] have collected 12 hours of transportation data in a controlled setting using only accelerometers. Yu et al [3] collected 8311 hours of transportation data with a single smartphone using only inertial sensors.…”