Real-Time Object Tracking with Template Tracking and Foreground Detection Network.

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Object tracking is crucial for various applications.
• Existing methods face challenges in accuracy and speed.
• Robust and efficient object tracking remains an active research area.

Purpose of the Study:

• To propose a fast and accurate deep network-based object tracking method.
• To integrate feature representation, template tracking, and foreground detection into a unified framework.
• To achieve robust tracking performance with real-time processing capabilities.

Main Methods:

• A deep network framework comprising a backbone network (modified VGG), a template tracking network (TmpNet), and a foreground detection network (FgNet).
• FgNet utilizes a fully convolutional network for pixel-wise foreground-background distinction.
• TmpNet employs a learned channel-wise target template for efficient tracking.
• A multi-task loss function enables end-to-end training of the integrated framework.

Main Results:

• The proposed method demonstrates favorable tracking accuracy compared to state-of-the-art trackers on benchmark datasets.
• The framework achieves a real-time tracking speed of 38 frames per second.
• Combining score maps from TmpNet and FgNet optimizes online tracking results.

Conclusions:

• The proposed deep network-based object tracking method offers a robust and efficient solution.
• The integrated framework effectively combines feature representation, template tracking, and foreground detection.
• The approach achieves a balance between high accuracy and real-time performance in object tracking.