Vehicle Counting System Using Image Processing with Pre-trained Machine Learning Models

Abstract

Understanding tourist flow dynamics is crucial for decision-making authorities in strategic tourism policy planning, infrastructure development, and natural resource management. This data enables informed decision-making to develop a sustainable tourism industry, promoting economic benefits while minimizing negative impacts on destinations. In this context, the city of Saltos del Guairá, Paraguay, whose economy is linked to the flow of tourists from Brazil, faces similar challenges due to the lack of a data collection mechanism to analyze incoming tourist flow. To address this challenge, the work proposes the use of a vehicle counting system based on image processing with pre-trained machine learning models for analyzing tourism flows in Saltos del Guairá. The methodology to achieve the expected results was divided into four stages. The first phase involved the evaluation of two machine learning models to select the one with the best performance in vehicle detection and tracking according to the requirements of this work. The second stage was the analysis and design of the system, with the aim of obtaining an architecture that allows for the interoperability and scalability of the system. In the third stage, the designed system was developed, using the selected machine learning model and the designs obtained in the previous phases. Finally, integration tests and the commissioning of the entire system were carried out. The proposed solution is valid for the problem, as it allows monitoring the flow of vehicles entering the city.