Visual Tracking Using Sparse Coding and Earth Mover's Distance.

Area of Science:

• Computer Vision
• Robotics
• Machine Learning

Background:

• Visual tracking is crucial for robotics and computer vision applications.
• Existing algorithms struggle with challenges like illumination variation, deformation, and rapid camera motion.
• Robustness in visual tracking is essential for real-world applications, especially on moving platforms.

Purpose of the Study:

• To propose an efficient iterative Earth Mover's Distance (iEMD) algorithm for visual tracking.
• To enhance tracking robustness against illumination changes, deformation, and large inter-frame displacements.
• To develop a gyro-aided extension for compensating camera rotation during rapid motion.

Main Methods:

• Utilized Earth Mover's Distance (EMD) as a similarity measure for template matching.
• Employed local sparse representation as the appearance model.
• Implemented maximum-alignment-pooling for efficient sparse coding histogram construction.
• Introduced a template update algorithm based on EMD.
• Integrated synchronized gyroscope data for camera rotation compensation in a gyro-aided iEMD tracker.

Main Results:

• The iEMD algorithm demonstrated robust performance against illumination variation and deformation.
• The algorithm proved effective in handling large inter-frame displacements.
• Experimental results on the CVPR 2013 dataset showed superior performance compared to eight state-of-the-art trackers.
• The gyro-aided extension successfully compensated for camera rotation, ensuring tracking convergence.

Conclusions:

• The proposed iEMD algorithm offers an efficient and robust solution for visual tracking.
• The integration of EMD and sparse representation enhances tracking accuracy and stability.
• The gyro-aided extension significantly improves tracking performance in dynamic environments with rapid camera motion.