IoT Enabled Traffic Signal Monitoring and Control System using Machine Learning

Abstract

The quality of traffic observations and the capacity of a controller to redistribute green time as demand changes are the key factors in adaptive traffic-signal control. This study proposes an Internet of Things (IoT) based traffic-monitoring architecture with MobileNet-SSD vehicle detection, approach-level queue estimation, and bounded queue-responsive signal timing and tests the timing policy in a reproducible four-approach point-queue simulation. The architecture is placed in the context of lightweight vehicle detection and vision-based traffic control, by reviewing ten empirical studies. Thirty paired replications are used for four demand scenarios and three timing policies, resulting in 360 simulation runs. In the case of switching demand, the queue-adaptive policy achieves a 59.41% decrease in mean total queue compared to a single demand-matched fixed-time policy with a demand of scenario-average demand, while the discharge increases by 1.84% compared to the single demand-matched fixed-time policy. The benefit is conditional: it grows by 15.01% in the balanced case, and by 1.27% in the stationary unequal case, but it does not seem to improve in the oversaturated case.