An adaptive spatiotemporal correlation filtering visual tracking method.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Discriminative Correlation Filter (DCF) trackers are widely used in visual tracking.
• Existing DCF trackers often fail to utilize spatio-temporal information, limiting their performance with targets of varying appearances.

Purpose of the Study:

• To develop an advanced tracker that effectively exploits spatio-temporal information for robust visual tracking.
• To address the limitations of current DCF trackers in handling diverse target appearances and motion patterns.

Main Methods:

• Proposed the Adaptive Spatio-Temporal Correlation Filtering (ASCF) tracker.
• Extracted rich local and global visual features using the Conformer network.
• Established and independently operated three correlation filters at different spatio-temporal locations.
• Employed Average Peak-to-Correlation Energy (APCE) and Peak-to-Sidelobe Ratio (PSR) for adaptive filter selection.
• Introduced an adaptive model update strategy to mitigate model drift.

Main Results:

• The ASCF tracker demonstrated superior performance across five benchmark datasets.
• The proposed method achieved excellent results compared to existing state-of-the-art trackers.
• The adaptive strategies for filter selection and model updating proved effective.

Conclusions:

• The ASCF tracker significantly improves visual tracking accuracy by integrating spatio-temporal information.
• The Conformer network-based feature extraction and adaptive filtering mechanisms contribute to robust performance.
• The algorithm offers a promising advancement in visual object tracking technology.