Deep Efficient Data Association for Multi-Object Tracking: Augmented with SSIM-Based Ambiguity Elimination.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Multiple Object Tracking (MOT) is crucial in video analysis.
• Existing MOT methods often rely on a two-step process: object detection and data association.
• Deep learning has shown promise in enhancing MOT performance.

Purpose of the Study:

• To propose an efficient and unified data association method for MOT.
• To improve the accuracy and robustness of multiple object tracking.
• To leverage deep learning for enhanced data association in MOT.

Main Methods:

• Developed a deep feature association network (deepFAN) for learning associations.
• Integrated the Structural Similarity Index Metric (SSIM) to handle data association uncertainties.
• Combined deep features and SSIM for effective linking of detections across frames.

Main Results:

• The proposed method demonstrated superior performance on standard MOT metrics.
• Achieved substantial improvements compared to current state-of-the-art MOT techniques.
• Validated through comprehensive analysis on MOT Challenge and UA-DETRAC datasets.

Conclusions:

• The proposed deepFAN and SSIM integration offers an effective MOT solution.
• The unified approach enhances tracking performance and accuracy.
• This work advances the state-of-the-art in deep learning-based multiple object tracking.