Building a pothole detection and tracking system
Building and maintaining infrastructure is often a key challenge in developing countries, and Ghana is no exception. Increasing population and car ownership rates coupled with poor maintenance cultures result in a corresponding increase in the rate of damage of roads, causing deformities such as cracks and potholes. These road deformities not only negatively impact a country’s road infrastructure and the cars which ply said roads, but also pose a threat to road users. In Ghana, only two mobile maintenance units are charged with monitoring the roads in all ten regions of the country. Thus, this project presents Pothole Tracker Ghana, a two-tiered application inspired by the idea of crowdsourcing. Consisting of a vision-based pothole classification system implemented on a Raspberry Pi and a map-based web application, this project aims to reduce the barriers to data collection on poor road infrastructure on the part of governments whilst allowing everyday road users to make informed decisions concerning their journeys. Three different algorithms are considered and compared for the classification task; logistic regression, support vector machines (SVM) and a hybrid algorithm incorporating a convolutional neural network (CNN) and SVM. The tuned SVM is chosen for the final system implementation.
Capstone Project submitted to the Department of Engineering, Ashesi University in partial fulfillment of the requirements for the award of Bachelor of Science degree in Computer Engineering, April 2019
Ghana, pothole detection, road classification, vision-based machine learning approach, Support Vector Machine (SVM)