Monitoring Road and Traffic Condition using Smartphone Devices

Abstract

Keeping roadways in good condition is challenging, because of normal wear and tear and unexpected traffic load. A ubiquitous monitoring system is required to inform drivers about unexpected road anomalies. Detecting these anomalies will lead to reducing traffic jams and road accidents. Most of the current road monitoring systems install dedicated sensors on the roadside or in dedicated vehicles such as inductive loops, traffic cameras, doppler radar, and laser sensors. These systems are expensive and infeasible to be ubiquitous since they are limited to the number of sensors deployed in the system. Smartphones have transformed from being basic devices into advanced tools comprising various communication channels, capable computational hardware, and reachable to a diverse set of sensors for interacting with the surroundings. Smartphones come with rich embedded sensors (e.g., GPS, accelerometer, and gyroscope) as well as built-in radios (e.g., Bluetooth, Wi-Fi, and Cellular), which both enable users to gather data and share them in the surrounding at any time or location. These capabilities enable the realization of mobile crowdsensing systems. In this thesis, the problem of monitoring road conditions using smartphone sensors is studied. We propose a complete mobile crowdsensing framework for road surface condition detection. In this framework, various modules along with typically utilized techniques and algorithms are discussed. The framework addresses various modules including task management, data fusion, reputation scoring, incentive awarding, security, and privacy. The proposed framework confirms the feasibility of utilizing the smartphone sensors in a real-time ubiquitous monitoring system for road surface conditions. Moreover, we focused on utilizing machine-learning techniques to detect road anomalies using smartphone sensors. An android application is developed to record sensor readings while driving the car. Using this app, a dataset of four trips on Cairo roads is constructed, consisting of a total duration of 80 minutes and about 50K records. To automatically label these records, two clustering techniques (K-Means and DBSCAN) are evaluated to identify the ground truth for the sensor readings if they represent road anomalies or normal road surfaces. It is noticed that DBSCAN can more accurately cluster sensor readings than K-Means can do. Once the dataset is labeled, a classification model is built to allow a smartphone to classify sensor readings and identify the surface quality of unseen roads. An accuracy of 96% can be obtained from the built classifier confirming the effectiveness of the adopted methodology in evaluating the road surface quality in Egyptian roads.