Traffic Management Using Image Recognition and Machine Learning

Abstract

Traffic lights, one of the methods of managing traffic, will greatly affect the waiting time of cars at junctions. This study aims to examine the performance of the intelligent traffic control system (ITCS) compared to the traditional and widely used fixed cycle traffic light (FCTL) control system in Malaysia. The object recognition model uses Tensorflow to identify the waiting cars at the junction. Simulation was done using Pygame with different car spawn rates at each junction during peak and non-peak hours. The simulation results show that the improvement in the average waiting time for cars is reduced when the spawn rate of cars increases. ITCS is capable of reducing the average waiting time of cars during non-peak hours by 25% or 3.6 seconds with a car spawn rate of 4 seconds at each junction. It has been identified that this ITCS design is capable of handling low traffic conditions very well. In peak hours, ITCS seems to struggle with the algorithm setting that was set during this study.