Research Article
RoadSpeedSense: Context-Aware Speed Profiling from Smart-phone Sensors
@ARTICLE{10.4108/eai.7-1-2020.162802, author={Ratna Mandal and Pallav Sonowal and Manish Kumar and Sujoy Saha and Subrata Nandi}, title={RoadSpeedSense: Context-Aware Speed Profiling from Smart-phone Sensors}, journal={EAI Endorsed Transactions on Energy Web}, volume={7}, number={28}, publisher={EAI}, journal_a={EW}, year={2020}, month={1}, keywords={Speed Profiling, Honk, WiFi, Road Condition, Intersection Density, Interpretable Machine Learning}, doi={10.4108/eai.7-1-2020.162802} }- Ratna Mandal
Pallav Sonowal
Manish Kumar
Sujoy Saha
Subrata Nandi
Year: 2020
RoadSpeedSense: Context-Aware Speed Profiling from Smart-phone Sensors
EW
EAI
DOI: 10.4108/eai.7-1-2020.162802
Abstract
INTRODUCTION: There are several online mapping systems like Google Maps, Waze, Here, Apple Maps, Bing Maps, etc. which are developed to visualize real-time traffic conditions which rely on crowdsourced GPS trails; obtained from worldwide smartphone users. Such systems still suffer from some limitations like a) inadequate traffic information in suburban cities and rural zones, b) system failure to infer the proper reasons for slow traffic state, c) difficulties in the extraction of raw traffic data for further development of any customized application. Significant spatio-temporal similarity patterns are observed in city traffic behavior unless there are some exceptional events like any disaster, VIP visit, international cricket match or bad weather condition, etc.
OBJECTIVES: Designing a framework and developing a system which enables collection of raw sensor information from users and to identify a model to generate a speed profile of city roads using historical logs as well as to infer the context of slow traffic based on ambient subjective road features and to provide map visualization.
METHODS: We have used road surface quality, density of vehicles, type of neighborhood and road geometry for developing speed profile for a particular road segment. We have carried out the experiments on different classification algorithms like, K-nearest Neighbor(KNN), Decision Tree(DT), Random Forest(RF) and Gradient Boost(GB) with necessary tuning of parameters.
RESULTS: GB outperforms other classification algorithms in estimating the speed class of road segments among all classifier algorithms with highest F1- Score of 0.8345. A fair driver rating system which can be derived from our results.
CONCLUSION: The results obtained from the proposed novel framework provide a proof of concept that speed profiles may be successfully derived from ambient road features even when sample space is sparse.
Copyright © 2020 Ratna Mandal et al., licensed to EAI. This is an open access article distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/3.0/), which permits unlimited use, distribution and reproduction in any medium so long as the original work is properly cited.
