Multicue HMM-UKF for real-time contour tracking.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Contour detection is crucial for object recognition and tracking.
• Existing methods often struggle with background clutter and nonlinear dynamics.
• Integrating multiple visual cues can improve detection accuracy.

Purpose of the Study:

• To develop an improved Hidden Markov Model (HMM) for contour detection.
• To enhance the model's robustness against background clutter.
• To enable robust contour tracking in nonlinear systems.

Main Methods:

• Proposed a HMM for contour detection utilizing multiple spatial domain visual cues.
• Implemented joint probabilistic matching to mitigate background clutter.
• Integrated an unscented Kalman filter for exploiting object dynamics.

Main Results:

• The proposed HMM demonstrates effective contour detection.
• Joint probabilistic matching significantly reduces background clutter.
• The integrated unscented Kalman filter ensures robust contour tracking.

Conclusions:

• The combined HMM, joint probabilistic matching, and unscented Kalman filter offer a robust solution for contour detection and tracking.
• This approach is particularly effective in nonlinear systems with significant background noise.