An Appearance-based Approach to Detect the Wrong-way Movement of Vehicles Using Deep Convolutional Neural Network

Abstract

To guarantee the enforcement of traffic rules, the identification of traffic rule violators is an exceptionally alluring yet difficult assignment to implement and the detection of the wrong-way movement of vehicles is one of them. In this paper, an appearance-based approach is proposed which detects the front and back side of the vehicles on a highway with the help of a deep convolutional neural network and decides whether a vehicle is moving along the wrong-way or not based on the user expectation to see the side of a vehicle on each side of the highway using a handcrafted region divider algorithm. The effectiveness of this strategy has been assessed on a primary data-set built on real-time traffic videos captured from several significantly busy highways of Dhaka Metropolitan City and proven quite productive with an accuracy of 96% on successful detection of wrong-way movement of vehicles.