Rastreamento de veículos em cidades inteligentes por meio de sensoriamento móvel participativo

Abstract

In the context of Smart Cities, vehicle tracking plays a fundamental role in applications such as urban mobility optimization, traffic management, and, especially, the enhancement of public safety. In this scenario, the identification and monitoring of stolen or stolen vehicles represent a significant challenge, particularly in dynamic urban environments with limited coverage of fixed infrastructure.This work proposes a low-cost vehicle tracking model based on Mobile Crowdsensing, in which videos of the road environment are captured by smartphones positioned in vehicles or at strategic inspection points. The model integrates computer vision and deep learning techniques, encompassing vehicle detection and tracking, extraction of license plate regions using the YOLO model, and character recognition using PaddleOCR. To increase reading reliability, the system automatically distinguishes between the Mercosur standard and the discontinued Brazilian format, applying specific validation rules for each case. Different processing strategies were experimentally evaluated, considering both remote processing approaches and local execution directly on mobile devices.The results indicate that local processing presents more consistent performance for the vehicle tracking task, achieving an average recall of 81.56% (recall, that is, the rate of correct identification of vehicles actually present in the video), in addition to greater statistical stability when compared to approaches dependent on data transmission. As the main finding of this research, local processing proved to be the most suitable strategy for the evaluated context, as it provides greater operational stability, lower dependence on connectivity, and feasibility for continuous field use. In contrast, remote approaches were more sensitive to network conditions. Additionally, this strategy operated within acceptable limits of computational resource consumption, reinforcing its practical applicability. The results demonstrate that the proposed model constitutes a robust and scalable alternative for vehicle tracking applications in Smart Cities, especially in supporting public safety actions, even in challenging dual mobility scenarios.