An Edge-AI Heterogeneous Solution for Real-time Parking Occupancy Detection

In the digital era, building smart cities is a highly desired goal that every country strives to achieve. With the advancement of technology, many smart city systems have been developed at a rapid rate of which Smart Parking is emerging as one of the core components. Smart Parking promises to automate the parking process, thereby saving time, resources and effort for searching an optimal parking space as well as reducing traffic congestion and population. As one of the newly emerging and disrupting technology, Artificial Intelligence, Machine Learning and Deep Learning (AI/ML/DL) are being utilized in many aspects of developing a Smart Parking system. In this paper, we propose an solution for accelerating AI/ML/DL algorithms deployed on low-cost System-on-Chip platforms (SoCs), which are often used as edge devices in Smart Parking system. In particular, we leverage Binary Neural Network (BNN), one of the most advanced deep learning models, to build a heterogeneous algorithm for real-time identifying parking occupancy based on the integration of SoCs and existing surveillance systems. The proposed solution is implemented and evaluated in Zynq UltraScale+ MPSoC with high accuracy (approx. 87%), low latency (avg. 16ms) and high frame per second (FPS) rate.

Automated summary

Building smart cities is a highly desired goal in the digital era, with Smart Parking emerging as a core component that promises to automate the parking process, save time and resources, reduce traffic congestion and population density. Utilizing AI/ML/DL algorithms on low-cost System-on-Chip platforms for real-time parking occupancy identification in Smart Parking systems shows promising results with high accuracy, low latency, and high frame per second rate.