Onboard Robust Visual Tracking for UAVs Using a Reliable Global-Local Object Model.

Area of Science:

• Computer Vision
• Robotics
• Aerospace Engineering

Background:

• Object tracking is crucial for Unmanned Aerial Vehicle (UAV) applications like autonomous navigation and target pursuit.
• Existing visual tracking algorithms often struggle with robustness and efficiency in real-world scenarios.

Purpose of the Study:

• To develop a novel onboard robust visual algorithm for long-term arbitrary 2D and 3D object tracking in UAV applications.
• To enhance the accuracy, efficiency, and robustness of object tracking for moving targets.

Main Methods:

• A global-local object model combining global feature matching with an improved binary descriptor and local feature tracking using the Lucas-Kanade optical flow algorithm.
• An efficient local geometric filter (LGF) employing a forward-backward pairwise dissimilarity measure and hierarchical agglomerative clustering for outlier feature correspondence handling.
• A heuristic local outlier factor method, applied for the first time in visual tracking, to classify outlier features and maximize target object representation.

Main Results:

• The proposed visual tracker operates at real-time frame rates exceeding 35 frames per second on an i7 processor with 640 × 512 resolution.
• The algorithm demonstrates superior performance compared to state-of-the-art trackers in terms of robustness, efficiency, and accuracy during extensive UAV flight experiments.

Conclusions:

• The novel global-local visual tracking algorithm provides a robust and efficient solution for UAV applications.
• The integration of advanced feature matching, geometric filtering, and outlier detection significantly improves tracking performance and reliability.