Motion-Aware Correlation Filters for Online Visual Tracking.

Area of Science:

• Computer Vision
• Artificial Intelligence

Background:

• Discriminative Correlation Filters (DCFs) methods face performance degradation due to fast motion and heavy occlusion in visual object tracking.
• These challenges often lead to tracking failures, necessitating improved tracking algorithms.

Purpose of the Study:

• To propose a novel Motion-Aware Correlation Filters (MACF) framework for robust online visual object tracking.
• To address the limitations of existing DCF methods in handling challenging motion and occlusion scenarios.

Main Methods:

• Integration of a motion-aware strategy using joint instantaneous motion estimation Kalman filters with DCFs.
• Prediction of target region and scale using estimated 3D motion information to mitigate model drift.
• An adaptive model updating strategy based on response map confidence to handle occlusions.

Main Results:

• The MACF tracker demonstrated superior performance compared to state-of-the-art trackers on OTB-100, OTB-50, and UAV datasets.
• Achieved high real-time performance, indicating practical applicability.
• The proposed approach showed flexibility for integration into other visual tracking algorithms.

Conclusions:

• The MACF framework effectively enhances online visual object tracking by incorporating motion awareness and adaptive occlusion handling.
• The method offers a significant improvement in tracking accuracy and robustness, especially under challenging conditions.
• MACF provides a valuable and adaptable solution for advanced visual tracking applications.