Bayraktar E.
2024 Innovations in Intelligent Systems and Applications Conference (ASYU), Ankara, Turkey, 16 - 18 October 2024, pp.1-6, (Full Text)
-
Publication Type:
Conference Paper / Full Text
-
Doi Number:
10.1109/asyu62119.2024.10757018
-
City:
Ankara
-
Country:
Turkey
-
Page Numbers:
pp.1-6
-
Yıldız Technical University Affiliated:
Yes
Abstract
This paper proposes
enhancements to state-of-the- art multi-object tracking methods,
focusing on improving pri- marily speed and then accuracy through
optimized matching algorithms. Specifically, we optimize the Kalman
Filter (KF), a foundational component in tracking frameworks principally
used as the motion model, by streamlining prediction and update
processes. Our approach includes efficient matrix operations and
explores the Unscented Kalman Filter (UKF) for handling nonlinear
dynamics robustly. Experimental results on MOT16, MOT17, and MOT20
benchmarks demonstrate a significant improvement in Multiple Object
Tracking Accuracy (MOTA), outperforming existing methods. While
achieving a MOTA of 65.7% on MOT17, our method shows promise in
mitigating occlusion and enhancing track continuity in complex scenes.
Future work includes further adaptation to diverse datasets and
exploring hybrid methodologies integrating deep learning for enhanced
performance in real-world applications.