Abstract: Moving object tracking is a key problem in computer vision and has many applications in various fields. This paper proposes a collaborative tracking method based on Co-Training frame work. The method fuses information from two types of features space to describe the object. The model is updated with Co-Training, which avoids the error accumulation problem. The experiment demonstrates the performance of the method under complex scenarios.

Key words: object tracking, Co-Training, semi-supervised learning, feature fusion

摘要： 运动目标跟踪是计算机视觉的核心问题之一，广泛应用于诸多领域。该文提出一种基于Co-Training半监督学习框架的目标跟踪方法。该方法融合2种互相独立的特征信息来描述目标模型，采用Co-Training来协同更新模型，有效避免了现有方法的误差累积问题。实验结果证明，该方法在复杂场景下仍能实现稳定有效的跟踪。

关键词: 目标跟踪, 联合训练, 半监督学习, 特征融合

CLC Number: 

• TP18