Contour-based object tracking with occlusion handling in video acquired using mobile cameras

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Object tracking is crucial for many applications, but challenges like changing appearance and occlusions persist.
• Existing methods often struggle with maintaining track under dynamic visual conditions and partial object visibility.

Purpose of the Study:

• To develop a robust object tracking method capable of handling appearance variations and occlusions.
• To improve the accuracy and reliability of object tracking in complex visual environments.

Main Methods:

• Object tracking via contour evolution, minimizing an energy functional within a defined band.
• Modeling visual features (color, texture) using semiparametric models and fusing them with independent opinion polling.
• Incorporating shape priors through shape level sets to recover occluded object regions.

Main Results:

• The proposed method successfully tracks complete object regions, adapting to changing visual features.
• Demonstrated effective handling of occlusions by recovering missing object parts using shape priors.
• Validated performance on real-world image sequences under various challenging conditions.

Conclusions:

• The developed contour evolution and shape prior method offers a robust solution for object tracking.
• The fusion of visual features and shape information enhances tracking stability and accuracy.
• This approach shows significant potential for real-time applications requiring reliable object tracking.