Authors - The Tung Than, Thi Phuong Nhi Le, Dong Thanh Vo, Minh Son Nguyen Abstract - Object tracking is crucial in computer vision, particularly in robotics, but Visual Object Tracking (VOT) faces significant challenges, with occlusion being the most critical. Occlusion disrupts tracking accuracy and poses difficulties when integrating VOT algorithms into embedded robotic systems due to computational and real-time constraints. To address this, we propose a robust method tailored for resource-limited systems, combining the Kernelized Correlation Filter (KCF) and Kalman Filter (KF). By leveraging the Average Peak-to-Correlation Energy (APCE) index, our method detects occlusion, dynamically adjusts the model’s learning rate, and improves performance under challenging conditions. Experimental results on the OTB-100 benchmark highlight our tracker’s effectiveness in handling occlusion, achieving a success rate of 0.602. This demonstrates the method’s robustness under challenging conditions while maintaining real-time processing at 30 FPS (Frame per Second) on Jetson Nano, making it an ideal solution for embedded robotic systems.
Open Zoom
