Vehicle Make and Model Recognition from Low-Detail Bangkok Metropolitan Administration CCTV Footage

Abstract

In this paper, we introduce a deep learning-based system specifically engineered for the recognition of vehicle makes and models under challenging conditions, including low-resolution image and variable lighting scenarios. To achieve efficient object detection, we employed the You Only Look Once (YOLO) architecture, which is renowned for its ability to quickly process images while maintaining high accuracy. Our system is adept at detecting full-body vehicles captured in surveillance footage from Bangkok Metropolitan Administration CCTV cameras. The current implementation includes well-known models such as the Honda City, Civic, and Accord, as well as the Toyota Vios, Camry, and Corolla Altis (Altis). To evaluate the system's effectiveness, we conducted experiments across multiple lanes, simulating various distances and viewpoints typical of real-world surveillance situations. Zone B lane 2 achieved the highest precision of 100% with F1score of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 9 9}$</tex>. Other zones also exhibited commendable performance metrics, achieving F1-scores between 0.90 and 0.96. These findings illustrate the robustness of our system and its considerable potential for practical applications.