Multi-Complementary Model for Long-Term Tracking.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Video target tracking algorithms are crucial but struggle with real-world challenges like occlusions, background clutter, motion blur, and illumination variations.
• Existing trackers often fail to maintain robust performance under these adverse conditions, limiting their practical applicability.

Purpose of the Study:

• To develop a more robust and efficient video target tracking algorithm capable of handling challenging scenarios.
• To improve tracking accuracy and speed by integrating complementary object models and optimizing detection mechanisms.

Main Methods:

• Incorporated an object model based on contour information into the Staple tracker, combining it with correlation filter and color models for complementary feature tracking.
• Proposed an efficient object detection model utilizing contour and color histogram features.
• Optimized traditional scale calculation for enhanced tracking execution speed.

Main Results:

• The enhanced tracker demonstrated significant improvements on the OTB-13 and OTB-15 benchmark datasets compared to the Long-term Correlation Tracking (LCT) algorithm.
• Achieved notable percentage increases in success plots for OPE, TRE, and SRE metrics on both datasets.
• The algorithm maintained good tracking speed due to the computational efficiency of its models and the speed advantage of correlation filters.

Conclusions:

• The proposed tracking algorithm significantly enhances robustness and accuracy in video target tracking, particularly in the presence of occlusions and illumination changes.
• The integration of contour and color models, along with an efficient detection mechanism, provides a powerful solution for challenging tracking tasks.
• The algorithm offers a favorable balance between tracking performance and computational efficiency, making it suitable for real-world applications.