Sistema inteligente de monitoramento de tráfego veicular com detecção adaptativa para condições diurnas e noturnas

Abstract

The continuous growth of vehicle traffic in urban areas poses significant challenges for mobility planning and management. Traditional monitoring approaches, such as sensors and conventional video systems, are often intrusive, high-cost, and lack the flexibility required for comprehensive real-time analysis. In response to this problem, this capstone project proposes the development of an intelligent vehicle traffic monitoring system using computer vision and machine learning techniques. The methodology employed is based on the YOLOv11 model for vehicle detection. To enhance the model’s robustness, a hybrid dataset was created by combining public images with a custom-collected dataset. One of the main challenges encountered was the variation in lighting, especially in nocturnal environments. To overcome this, the frozen layers technique was applied, allowing for the fine-tuning of the model with a smaller training dataset. Ultimately, this work demonstrates the feasibility of applying lightweight machine learning models for traffic monitoring, even with limited computational resources. Furthermore, it was shown that the frozen layers technique is an effective approach for adapting models to low-light conditions. The application of this methodology enabled the generation of a detailed analysis of vehicle flow, providing valuable data for urban planning. The results obtained prove the effectiveness of the proposed system, showing satisfactory performance in both daytime and nighttime conditions.