Abstract:
With the continuous increase in population and drastic urbanization trends, roads have become a key perfonnance indicator of a city's smooth running. Road safety is an important concern for every city; surface conditions and quality of roads have significant share in safety and smooth driving experience. Smooth roads tend to reduce the traffic congestion while ensuring maximum safety of the passengers (as well as the vehicles). In Pakistan, due to lack of proper road maintenance, many fatal incidents are recorded on daily basis. However, with the technological revolutions, there has to be an automated solution for precise monito1i ng of roads. Due to scarcity of relevant scientific data for this type of study, a systematic data collection approach was adapted to acquire the road readings through a custom-bui lt smartphone application. Sensors (GPS, Accelerometer and Gyros) on drivers' smartphones record spatiotemporal readings al sufficient rates. An algorithm is proposed that takes the collected data, optimises it to identify and prominent the road anomalies from it, and extract features by applying feature extraction techniques. A supervised machine learning approach then generates relevant road models for different anomalies like potholes, speed breakers, man holes etc. The research also includes a study of a multiple machine learning classifiers and their capabilities in detecting major irregularities on the basis of the road surface data. For this research, some of the busiest roads of Islamabad were monitored using an application which is installed on drivers' smartphones. The application grabs the inertial readings and transmits the data to a remote server that utilizes various Data Mining and Machine Learning schemes for generating road models and anomaly classes. The resultant road model will be used to classify a candidate anomaly and will indicate its severity based on its prominent features. The results showed a significant accuracy in identifying the anomalies based on their severity. A table from 1-5 was used to label the road anomalies based on the discomfort and unintended behavior they produced for the the commuters. This research can be further enhanced to a variety of application including driver warning systems, road surface monitoring systems etc.
