Enhancing Real-Time License Plate Recognition Through Edge-Cloud Computing

Video makes up a significant portion of data gathered from the edge [1]. A notable source of video data is the dashboard camera, and modern models can now be connected to the Internet of Things (IoT). With this video data, automatic license plate recognition (ALPR) algorithms can be performed to locate vehicles, which can be used in investigations such as child abductions and vehicle theft. However, this volume of video data only increases with the rising number of connected devices, putting a large strain on the network. Edge computing can help alleviate this strain on the bandwidth. We propose an edge computing-based ALPR system that locates a target vehicle given a video feed and GPS data. Specifically, we implement this system on a cloud server and a Raspberry Pi 4 as its edge. Comparisons between edge-heavy, cloud-heavy, and hybridized setups are done to evaluate its performance. We also evaluated their scalability and their performance in low-bandwidth conditions. Experimental findings show that edge-heavy and hybridized setups can scale up easily and can perform effectively in low bandwidth conditions (10 Kbps). Meanwhile, the cloud-heavy setup performs the best with a single edge, but performance suffers in low-bandwidth conditions and has poor scalability.