HKSiamFC: Visual-Tracking Framework Using Prior Information Provided by Staple and Kalman Filter.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Convolutional Neural Network (CNN) trackers have advanced visual tracking.
• The Fully-Convolutional Siamese (SiamFC) tracker, a prominent CNN tracker, excels at similarity learning but struggles with robustness in complex scenarios due to insufficient target prior information.

Purpose of the Study:

• To develop a novel and robust visual tracking framework that addresses the limitations of existing methods.
• To enhance the SiamFC tracker by integrating prior target information for improved accuracy and stability.

Main Methods:

• A Histogram-Kalman SiamFC (HKSiamFC) framework was designed, building upon the SiamFC architecture.
• Two supplementary models were developed: one leveraging target prior color information (histogram) and another utilizing target prior trajectory information (Kalman filter).

Main Results:

• The HKSiamFC tracker demonstrated competitive performance compared to the baseline SiamFC and other state-of-the-art trackers.
• Evaluations on the Online Object Tracking Benchmark (OTB) and Temple Color (TC128) datasets confirmed the tracker's robustness and effectiveness.

Conclusions:

• The proposed HKSiamFC framework significantly improves visual tracking robustness by incorporating color and trajectory priors.
• HKSiamFC offers a promising advancement in CNN-based visual tracking, particularly for challenging environments.