A Robust Tracking-by-Detection Algorithm Using Adaptive Accumulated Frame Differencing and Corner Features

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Robotics

Background:

• Overhead cameras offer clear views but present detection challenges due to wide-angle distortion and weak features.
• Standard motion detection methods fail without background models and struggle with diverse movement patterns common in meetings.
• Existing tracking algorithms often falter in cluttered environments with poor object distinctiveness.

Purpose of the Study:

• To develop a robust tracking-by-detection algorithm for monitoring meeting participants from an overhead perspective.
• To address the limitations of conventional methods in handling wide-angle distortions, weak features, and dynamic movement behaviors.
• To enhance the accuracy and reliability of multi-object tracking in challenging real-world scenarios.

Main Methods:

• A novel coarse-to-fine detection and tracking approach combining Adaptive Accumulated Frame Differencing (AAFD) for motion detection.
• Integration of Shi-Tomasi corner detection to identify salient features for robust tracking.
• Utilized experimental datasets without empty room priors and the Online Tracking Benchmark (OTB) for evaluation.

Main Results:

• Demonstrated robustness in tracking individuals with unclear features and background similarity.
• Achieved excellent performance on Multiple Object Tracking Accuracy (MOTA) metrics.
• Showcased superior robustness to initialization differences compared to baseline and state-of-the-art trackers.
• Validated strong performance against background clutter, deformation, and illumination variations using OTB videos.

Conclusions:

• The proposed algorithm offers a reliable solution for tracking people in challenging overhead camera views.
• The coarse-to-fine approach effectively overcomes limitations of traditional motion-based and feature-based tracking methods.
• This work contributes a robust and accurate tracking system suitable for meeting analysis and other applications.