Debaleen Das Spandan

In this work, we proposed a machine-learning driven context-aware framework entitled BuStop which can detect diferent types of stay-locations of a public bus, namely regular bus stop, stop at traic signal, stop due to excessive traffic congestion, stops due to turns on the road and finally the randomly given ad-hoc stops. The framework does this by correctly identifying and choosing context-aware features extracted from multiple modalities that allow the framework to discern between these stay-locations. Rigorous evaluation of the framework on the in-house collected dataset shows appreciable accuracy, thus providing an eicient way of characterizing the stay-locations. Additionally, we also develop a PoC system on top of the developed framework to analyze and identify the framework’s potential in providing an accurate expected time of arrival, one of the most critical pieces of information required for pre-planning the travel. Further analysis of the PoC setup, with simulation over the test dataset, shows that the stay-locations’ characterization allows the setup to predict the arrival time with a deviation of less than 60 seconds.

I worked on this project during my research internship at National Institute of Technology, Durgapur in 2020.This work has been published in ACM TIoT 2022.

Video showcasing Proof-of-Concept system