Computer Vision Methods for Vehicle Detection and Tracking: A Systematic Review and Meta-Analysis

Automatic vehicle detection and tracking are at the core of the latest smart city developments, enhancing mobility services across the globe. Nevertheless, research in this field often suffers from inconsistent results caused by heterogeneity in datasets, methodologies and evaluation metrics. These challenges highlight the need for this systematic review, which comprises the work of 29 peer-reviewed studies extracted from Scopus and ACM Digital Library published between 2020 and 2024, focusing on integrated vehicle detection–tracking systems using fixed top-down imagery. The selected works were critically examined according to their algorithms, methodological practices, dataset characteristics and performance metrics, culminating in a meta-analysis to quantify and fairly compare results. In parallel, the broader ecosystem surrounding vehicle detection and tracking was also explored to provide a complementary perspective, including evaluation standards and dataset diversity, helping to guide future works. The findings reveal that state-of-the-art research lacks standardization of metrics and reporting, heavily relies on datasets that are incompatible with tracking benchmarks and often limited in scenario diversity, and repeatedly exhibit methodological lenience compromising reproducibility and transparency. While the meta-analysis helps contextualize the best-reported implementations, the absence of standardized practices ultimately fragments the experiment. This review consolidates the current knowledge and suggests concrete directions to improve robustness, comparability and deployment of vehicle detection and tracking systems for future smart-cities infrastructures.