Gary Fernando Yunganina Mamani, Guver Leon Cori Coarite, Jhon Alexander Chambi Vilca, Ángel Rosendo Condori-Coaquira, David Mamani-Pari, Milton Edward Humpiri-Flores, Esteban Tocto-Cano
YOLO-based architectures were the most frequently adopted model family, appearing in seven of the 15 studies (46.7%). Selected studies reported mean average precision values above 90% and processing speeds above 40 FPS; however, cross-study comparisons were limited by heterogeneous datasets, hardware configurations, metric definitions, and experimental conditions. Environmental evaluation was limited and uneven: fog was examined in one of 15 studies, occlusion in two, and snow in none. PKLot and CNRPark-EXT were the most frequently used public benchmarks, whereas the use of private and mixed datasets limited external validation when data or metadata were not shared.
INTRODUCTION: Automated parking-space detection is an increasingly important component of intelligent urban mobility because inefficient parking searches contribute to travel delays, congestion, fuel consumption, and emissions.
METHODS: This systematic literature review, conducted using Kitchenham's methodology and the PICOC framework, synthesizes 15 peer-reviewed studies published between 2021 and March 2026. It examines detection methods, reported accuracy and computational efficiency, environmental evaluation coverage, dataset use, preprocessing practices, reproducibility, and future research directions.
RESULTS: YOLO-based architectures were the most frequently adopted model family, appearing in seven of the 15 studies (46.7%). Selected studies reported mean average precision values above 90% and processing speeds above 40 FPS; however, cross-study comparisons were limited by heterogeneous datasets, hardware configurations, metric definitions, and experimental conditions. Environmental evaluation was limited and uneven: fog was examined in one of 15 studies, occlusion in two, and snow in none. PKLot and CNRPark-EXT were the most frequently used public benchmarks, whereas the use of private and mixed datasets limited external validation when data or metadata were not shared.
DISCUSSION: This review provides an analytical performance matrix and a quantified environmental evaluation-coverage framework that identify recurring gaps in performance reporting, environmental validation, and reproducibility. The findings demonstrate the need for standardized comparative benchmarks, environmentally diverse datasets, and transparent reporting practices to enable robust assessment of real-world performance, robustness, and scalability.