The paper introduces a new color-based probabilistic tracking technique that enhances the robustness and versatility of existing deterministic color-based trackers. The proposed method uses a Monte Carlo framework, specifically a particle filter, to better handle color clutter and complete occlusions. This approach allows the tracker to maintain track of multiple posterior modes, enabling it to escape from background distractions and recover from partial or complete occlusions. The probabilistic tracking method is flexible and can be extended in several ways: 1. **Multi-part Color Modeling**: This extension captures the spatial layout of different color patches within a tracked region, improving tracking accuracy and stability. 2. **Background Modeling**: Incorporating a background color model enhances tracking robustness, especially in scenarios with fixed cameras and background knowledge. 3. **Multiple Objects**: The method can track multiple objects simultaneously, even when they overlap, by marginalizing over possible depth orderings. 4. **Automatic Initialization on Skin**: The tracker can automatically initialize on skin regions, making it suitable for face tracking in still scenes. The paper demonstrates the effectiveness of the proposed method through various experiments, showing improved performance over deterministic color-based trackers in handling complex scenes and occlusions. The authors also discuss connections between their method and the "Bramble" tracker, highlighting differences in foreground and background modeling and object handling.The paper introduces a new color-based probabilistic tracking technique that enhances the robustness and versatility of existing deterministic color-based trackers. The proposed method uses a Monte Carlo framework, specifically a particle filter, to better handle color clutter and complete occlusions. This approach allows the tracker to maintain track of multiple posterior modes, enabling it to escape from background distractions and recover from partial or complete occlusions. The probabilistic tracking method is flexible and can be extended in several ways: 1. **Multi-part Color Modeling**: This extension captures the spatial layout of different color patches within a tracked region, improving tracking accuracy and stability. 2. **Background Modeling**: Incorporating a background color model enhances tracking robustness, especially in scenarios with fixed cameras and background knowledge. 3. **Multiple Objects**: The method can track multiple objects simultaneously, even when they overlap, by marginalizing over possible depth orderings. 4. **Automatic Initialization on Skin**: The tracker can automatically initialize on skin regions, making it suitable for face tracking in still scenes. The paper demonstrates the effectiveness of the proposed method through various experiments, showing improved performance over deterministic color-based trackers in handling complex scenes and occlusions. The authors also discuss connections between their method and the "Bramble" tracker, highlighting differences in foreground and background modeling and object handling.