A Reinforcement Learning-based Offload Decision Model (RL-OLD) for Vehicle Number Plate Detection

Vehicle license number plate detection is essential for road safety and traffic management. Many existing systems have been proposed to achieve high detection precision without optimization of computer resources. Existing models have not preferred to use devices like smartphones or surveillance cameras because of high latency, data loss, bandwidth costs, and privacy. In this article, we propose a model of unloading decisions based on reinforcement learning (RL-OLD) for recognition and detection of vehicle license plates for high precision with optimization of computer resources. The proposed model detected different categories of vehicle registration plates by effectively utilizing edge computing. Our model can choose either the compute-intensive model of the cloud or the lightweight model of the local system based on the properties of the number plate. This approach has achieved high accuracy, limited data loss, and limited latency.