Edge AI-Enabled Smart Parking Systems

In recent years, edge computing has emerged as a promising paradigm for data computation, allowing for pre-processing of data prior to transmission to servers. This approach finds significant application in the development of intelligent systems for various domains such as urban environments, workplaces, and agriculture. In the context of smart cities, the adoption of smart parking systems has been on the rise, owing to their manifold advantages. Unlike conventional systems that necessitate considerable time investment from drivers in locating parking spaces, the smart parking approach with the support of edge AI platforms and cloud-based services help users to locate their booked slots quickly by utilizing video and sensor data. This paper introduce our proposed our smart parking system based on edge AI computing paradigm where AI models and heavy workload are processed by edge devices before sending data to cloud services. Our proposed architecture employs edge computing to handle the bulk of video processing tasks, thereby mitigating challenges related to network bandwidth. A prototype of our system developed with two Jetson Xaiver boards and two Yolo-based AI models have been deployed and validated inside our institution's campus Our experiments demonstrate that in terms of processing throughput, we obtains up to 30 frames per second (FPS) when detecting license plates and up to 24 FPS when recognizing empty parking slots. Furthermore, the AI models after training with certified datasets, achieve an accuracy and precision of up to 98.0% and 99.6%, respectively.