Intelligent Bus Stop Identification Using Smartphone Sensors

Abstract

Intelligent transportation systems can be built by developing models that learn from the collected transport data. Data collection and implementation of such systems is often costly, and few countries have support for such systems in their transportation budgets. In places where maintaining currency and accuracy of information is difficult, many problems arise. For instance, in Chennai, India, real time bus transit data is not maintained, there is no proper communication about the bus schedules, bus stops are not regularly updated and inconsistent information about bus stops is observed in the transport authority's website. We are interested in developing models for identifying bus stops from trajectories for situations where accurate and current information is not available and traffic conditions are challenging, such as Chennai, India. We develop a simple yet easily accessible Android mobile application (App) to collect GPS traces of bus routes. We use our App to collect GPS trajectory data from Baltimore, Maryland, a place where there are facilities to access up-to-date information about bus stops. We also collect GPS trajectories from Chennai, India. We then develop a model using machine learning techniques to identify bus stops from the collected trajectories. We experimentally evaluate our model by training it on the Baltimore dataset and testing it on the Chennai dataset, achieving testing accuracy between 85 -- 90%. This is comparable to the accuracy of 95% achieved by both training and testing on the Chennai dataset. This illustrates that our approach is effective in helping maintain an accurate and current transport information system for resource constraint environments.