Vehicle detection, tracking, counting and classification using deep learning

Abstract

This paper explores vehicle counting system, which is a crucial component of building a robust transportation management system (TMS) that proffers solution to various challenges facing transportation systems in modern cities around the world. Although there are existing approaches such as the manual vehicular counting and hardware-based systems but they are plagued with various limitations such as being intrusive on roads, difficult to scale and also very expensive to maintain. Hence, there are not enough viable solutions to the current complex and diverse traffic challenges. This paper focused on the development of a vehicle counting system designed to capture and read video in real-time from a camera placed strategically to capture traffic scenes and thereafter counts and classify the vehicles as they cross a detection line. A visualization of the results is displayed onscreen in real-time and the count data for all vehicle classes are saved in a database for future analysis. The counting and classification obtained accuracy is greater than 80%. This research achieved a software-based video counting system that runs on computer vision algorithms and presents an accurate, inexpensive, flexible, scalable and non-intrusive approach to obtaining vehicle count on highways.