中文摘要：

在TLD（TrackingLearningDetector）算法的基础上，提出了一种基于OnlineMIL（OnlineMultipleInstanceLearning）的TLD目标跟踪算法。算法使用改进的MIL跟踪器进行目标跟踪，提高了目标在被遮挡情况下跟踪的鲁棒性。另外，在学习机制部分，目标位置的决策策略使用了跟踪结果优先的原则，使算法能适应出现类似目标时的跟踪。实验结果表明，该方法能够长时间准确地跟踪目标，并在出现类似目标时；跟踪效果较好。此外，改进后的算法在跟踪的稳定性和跟踪效率上较原算法提高了1倍。

英文摘要：

On the basis of TLD algorithm,t a TLD method based on online MIL algorithm is proposed. Using the improved MIL tracking algorithm to improve the robustness in object tracking through occlusion. And in the part of learning, in order to tracking a siMILar target, this paper changes the decision strategies of the target location, and uses the principle of priority of tracking results to decide the target location. The experimental results show that the method can be long and accurate tracking targets and better tracking siMILar goals appear. In addition,the improved algorithm in tracking stability and tracking efficiency increase one times compared with the original algorithm.