Adaptive Traffic Control With TinyML

Automatic traffic scheduling can improve the efficiency of the traffic system by decreasing significant delays and easing congestion. This paper introduces a machine learning (ML) based algorithm to automate the traffic scheduling based on the density of vehicles waiting, without any human intervention. Presently, the traffic signal timers are preset which is unreasonable for a stochastic process like vehicular traffic. The proposed methodology detects vehicles with piezoelectric sensors embedded across each lane. The two-point time ratio method is utilized to identify the vehicles using the sensor's data and the vehicles are classified based on their pick-up speed. Further, a TinyML based model is proposed to predict the green signal timings.