Object Detection in Real-World Smart City Applications: A Case of Truck Detection in California Highways
Bhavyesh Sajja, Wenting Qiu, Jooyoung Yoo, Seon Ho Kim · 2025
Monitoring and analyzing vehicle movements, particularly those of freight trucks, is essential for traffic management, urban planning, and logistical efficiency. In the United States, trucks play a vital role in transportation, thus necessitating accurate and automated truck movement monitoring systems, especially under challenging conditions such as night-time, adverse weather conditions, or noisy data. This paper presents a comprehensive study of truck detection in real-world CCTV video footage of California Highways using a state-of-the-art object detection model. A large dataset sourced from CalTrans, covering 9 different CCTV locations and totaling 1.5 TB of video data, was analyzed. This analysis investigated the various challenges in truck detection under diverse real-world conditions. Experimental results demonstrate that in addition to noisy conditions in images, other factors such as camera angle and weather conditions play a key role in high-quality data analysis. Furthermore, we introduce an inpainting- based image preprocessing technique to enhance the robustness of truck detection against noises, particularly under low-light conditions, improving detection accuracy, and contributing to the development of more reliable and scalable truck traffic monitoring systems.