Bridging Research Gaps in Smart Parking Systems: A Comparative Study of CNN and VANET Approaches

Abstract

The combination of increased urbanisation and rapid vehicle diversification has made smart parking management an essential infrastructure tool for modern cities. In spite of three decades of research, current solutions face many limitations. Per-bay infrastructure costs for sensor-based systems become cost prohibitive. The accuracy of camera-based systems, using Convolutional Neural Networks (CNNs), is degraded by 5–23% when deployed across different domains. Communication frameworks based on Vehicular Ad-hoc Networks (VANETs) assume ideal conditions of sensing, overlook parking, and omit protection against manipulation of parking data. This paper provides a systematic review that aims to explain the different approaches of CNNs and VANETs for smart parking by integrating the approaches and highlighting research gaps where both paradigms must be accommodated. Based on a systematic review of 41 papers from the fields of urban informatics, edge computing, and other related fields, the authors present an integrated architectural framework that incorporates CNNs and the VANETs. The authors present a framework that includes domain-adaptive deep learning, a hierarchy of edge-fog-cloud computing, intrusion detection, and a federated learning framework to the context of parking. Six hypothesis are suggested out of the framework. The authors argue that integrating CNNs and VANETs is a necessary first step for smart urban parking of a large-scale system. The smart parking system is aimed at being highly scalable, secure, and privacy-oriented.