V3I5P30

A Deep Neural Network Based Traffic Control System in Intelligent Transportation System with Special Considerations for Emergency Vehicles and traffic density

Nwonye Charles A^1*, Onyia Tochukwu. C², Nweezekwunife Ozioma. G³

Abstract

This work proposes the development of a deep neural network based traffic control system in intelligent transportation system with special considerations for emergency vehicles and traffic density. This work would ensure that a deep neural network in the form of convolutional neural network is applied in traffic control system so as to ensure that traffic is managed in a more intelligent manner. Here, traffic is assigned to a particular direction based on the traffic density and the emergency situations. A you only look once version 4 model (YOLOV4) was used to identify the cluster of traffic in all the directions on the road. To avoid allocating right of passage to a direction without traffic and allocating very small time of passage to a direction with heavy traffic, the yolov4 model was used to determine the direction with the highest cluster of traffic and a right of passage is given to it. A direction with a right of passage continues until another direction becomes the highest cluster of traffic or an emergency vehicle appears in another direction. The system allows a pedestrian passage of 20 seconds after each transition. In this study, 200 images of normal traffic were captured by cameras and 50 images of traffic with emergency vehicles like ambulances, fire trucks, military and police vehicles were equally captured by cameras. Then, 2000 images of normal traffic and 100 images of traffic with emergency vehicles were also downloaded from the internet. Hence, total of 2350 traffic images were preprocessed using python language to form the training data set that was used to train the deep neural network ( yolov4) model in google colab platform using python language programming. The trained deep neural network was tested for classification of traffic into normal and emergency traffic using 700 traffic images and it was able to show precision of 96%, accuracy of 97%, recall of 0. 96 and F1score of 0.96. This model would reduce the wastage of traffic time in wrong direction as common with the traditional traffic control that even assigns traffic passage to a direction without any traffic on it. Hence, it would assign traffic passage based on the traffic density and emergencies on the road.

Keywords:

Intelligent Transportation System; Deep Neural Network; YOLOv4; Traffic Density Management; Emergency Vehicle Detection