Autonomous On-Street Parking Slot Detection: A Comprehensive Systematic Review of Vision-Based Approaches

Abstract

Finding an available parking space remains a significant challenge, particularly in major international cities, where rapid population growth and increasing traffic congestion exacerbate this problem. To address this issue, numerous sensor-based systems have been developed to manage parking infrastructures. However, these systems are often expensive and impractical because their sensors are susceptible to damage from adverse weather conditions. Consequently, in-vehicle camera-based parking-slot detection methods have emerged as more effective and scalable alternatives. Rapid advances in artificial intelligence in recent decades have facilitated the development of diverse methods and tools for autonomous parking detection. This systematic review examines state-of-the-art parking detection techniques, including image processing, machine learning, deep learning, and reinforcement learning approaches. It considers various imaging perspectives such as the driver’s point of view, bird’s-eye view, and around-view monitoring (AVM). The reviewed methods are evaluated using widely adopted benchmark datasets, with a focus on their performance in identifying parking slots and determining the occupancy status. Key challenges, including real-time implementation constraints and performance robustness across diverse environmental conditions, were critically analyzed. The findings suggest that hybrid models that integrate reinforcement learning with supervised learning hold substantial promise for future research and applications. This study offers a comprehensive assessment of current methodologies, identifies persistent research gaps, and proposes actionable recommendations to advance the development of robust autonomous parking detection systems.