Publications Repository - Gdańsk University of Technology

This paper introduces a Smart City solution designed to run on edge devices, leveraging NVIDIA's DeepStream SDK for efficient urban surveillance. We evaluate five object-tracking approaches, using YOLO as the baseline detector and integrating three Nvidia DeepStream trackers: IOU, NvSORT, and NvDCF. Additionally, we propose a custom tracker based on Optical Flow and Kalman filtering. The presented approach combines advanced machine learning and deep learning techniques to enhance object tracking in intelligent traffic management systems, contributing to the evolving landscape of urbanization. Experimental results highlight the challenges and potential improvements in tracking accuracy, particularly in addressing object misclassification. In the conducted study, the proposed method achieved average precision = 0.95.

Authors

• dr inż. Tomasz Kocejko link open in new tab ,
• dr Tomasz Neumann link open in new tab ,
• dr inż. Magdalena Mazur-Milecka link open in new tab ,
• mgr inż. Natalia Kowalczyk link open in new tab ,
• prof. dr hab. inż. Jacek Rumiński link open in new tab ,
• Jo Kang-Hyun,
• Miłosz Kaszyński,
• Tomasz Ludwisiak