Active contour-based visual tracking by integrating colors, shapes, and motions

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Visual tracking is crucial for analyzing dynamic scenes.
• Existing methods struggle with non-periodic, periodic, and abrupt object motions.
• Active contours and level sets offer a promising approach for object boundary delineation.

Purpose of the Study:

• To present a comprehensive framework for active contour-based visual tracking using level sets.
• To improve tracking robustness and accuracy across diverse motion patterns.
• To address limitations in contour initialization, evolution, and handling of motion dynamics.

Main Methods:

• Optical flow-based algorithm for automatic contour initialization.
• Markov Random Field (MRF) theory for color-based contour evolution.
• Hierarchical contour evolution combining global shape and local color information.
• Flexible shape updating models for non-periodic and periodic motions.
• Particle Swarm Optimization (PSO) for handling abrupt motions.

Main Results:

• Successfully initialized contours automatically using optical flow.
• Demonstrated robust color-based contour evolution via MRF posterior probability estimation.
• Achieved adaptive and dynamic shape-based evolution for non-periodic and periodic motions.
• Effectively handled abrupt motions using PSO for improved contour initialization in subsequent frames.

Conclusions:

• The proposed framework provides a robust and versatile solution for visual tracking.
• Integration of shape, color, and motion handling significantly enhances tracking performance.
• The framework offers a significant advancement in active contour-based visual tracking methodologies.