IoT-based traffic light control

Abstract

Machine learning is the process of teaching a set of artificial neurons to perform a task that is usually too difficult to perform programmatically. Traffic light systems are an example of such difficult tasks. The aim of this thesis is, therefore, to combine machine learning algorithms with the unlimited processing power of the cloud for use in traffic light control. The cloud/IoT aspect of this thesis reliefs computations from local traffic light controllers, thereby reducing hardware cost, and allows sensors and other components to function independently from one another. This reduces wiring and minimises intrusion upon installation onto existing traffic networks. The designed sensors and traffic controllers feature vehicle classification and the capability to upload real-time traffic information to the cloud. The developed reinforcement learning algorithm reduced vehicle wait times by an average of 29.3 per cent.