Advanced Kalman Filter Optimization for Efficient Multi-Object Tracking in Computer Vision

2024 Innovations in Intelligent Systems and Applications Conference (ASYU), Ankara, Türkiye, 16 - 18 Ekim 2024, ss.1-6

Yayın Türü: Bildiri / Tam Metin Bildiri
Doi Numarası: 10.1109/asyu62119.2024.10757018
Basıldığı Şehir: Ankara
Basıldığı Ülke: Türkiye
Sayfa Sayıları: ss.1-6
Yıldız Teknik Üniversitesi Adresli: Evet

Özet

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.