This paper presents a novel on-line semi-supervised boosting method for robust tracking, which significantly alleviates the drifting problem in tracking applications. The main idea is to formulate the update process in a semi-supervised fashion as a combined decision of a given prior and an on-line classifier. This approach allows the tracker to adapt to appearance changes while limiting drifting. The method uses labeled data as a prior and unlabeled samples collected during tracking. This enables a natural formulation of the tracker update problem without parameter tuning. The proposed SemiBoost tracker outperforms other on-line tracking methods in real-time tracking on challenging test sequences. The paper also discusses the limitations of traditional tracking methods, such as fixed classifiers that cannot adapt to appearance changes, and the drifting problem caused by on-line adaptation. The SemiBoost method addresses these issues by incorporating prior knowledge and unlabeled data, allowing the tracker to adapt to changes while maintaining robustness. The method is evaluated on various scenarios, demonstrating its effectiveness in handling different objects and appearance variations. The results show that the SemiBoost tracker can recover from tracking failures and maintain accurate tracking even in challenging conditions. The paper concludes that the proposed method provides a balance between adaptability and robustness, making it suitable for real-world tracking applications.This paper presents a novel on-line semi-supervised boosting method for robust tracking, which significantly alleviates the drifting problem in tracking applications. The main idea is to formulate the update process in a semi-supervised fashion as a combined decision of a given prior and an on-line classifier. This approach allows the tracker to adapt to appearance changes while limiting drifting. The method uses labeled data as a prior and unlabeled samples collected during tracking. This enables a natural formulation of the tracker update problem without parameter tuning. The proposed SemiBoost tracker outperforms other on-line tracking methods in real-time tracking on challenging test sequences. The paper also discusses the limitations of traditional tracking methods, such as fixed classifiers that cannot adapt to appearance changes, and the drifting problem caused by on-line adaptation. The SemiBoost method addresses these issues by incorporating prior knowledge and unlabeled data, allowing the tracker to adapt to changes while maintaining robustness. The method is evaluated on various scenarios, demonstrating its effectiveness in handling different objects and appearance variations. The results show that the SemiBoost tracker can recover from tracking failures and maintain accurate tracking even in challenging conditions. The paper concludes that the proposed method provides a balance between adaptability and robustness, making it suitable for real-world tracking applications.