SMART Emergency Vehicle Management at Signalized Intersection using Machine Learning

Abstract

Objective: Primary objective of the article is to develop a machine-learningbased pre-emptive traffic signal controller to ensure the free flow of an emergency vehicle across the signalized intersection. Method: Pre-emptive signal control system involves various functional modules such as emergency vehicle notification, traffic volume estimation, green time prediction, and signal control. The current study is focused on green time prediction based on traffic composition and volume. The study is presented in two folds; identify a suitable machine learning model to predict the green time and use the selected model to design the proposed system. The results obtained from the proposed system are compared against a non-pre-emptive controller. Findings: The Convolution Neural Network (CNN) is found to be the best suitable algorithm for green time prediction. The green time prediction module shares a pivotal role in the system as more/less green time prediction can waste the green time or block the free flow of emergency vehicles. Thus, the accuracy of green time prediction has significance in the system and CNN showed a 96% R2-score. Delay for an emergency vehicle is 86 seconds in a conventional non-pre-emptive controller and it is 8 seconds in the case of proposed system. Novelty: Green time prediction under heterogeneous traffic conditions is a challenge. Analytical models are widely used to estimate the green time as per existing research works concerned with emergency vehicle management. However, machine learning models are also in use, but deep learning models are applied rarely, and CNN is applied in current work. Keywords: Signal preemption; edge computing; machine learning; signalized intersection; emergency management