基于多特征自适应融合的目标跟踪算法

针对目标跟踪中部分遮挡及漂移问题,提出了一种基于多特征自适应融合和在线学习的目标跟踪算法。首先针对原始特征提取的相对简易性,提出了改进的特征提取方法,提高了特征的表达能力和判别能力,然后将多种特征自适应融合,最后进行在线学习。实验采用了具有挑战性的公共测试数据集PETS 2012,并用MOTP评测了跟踪性能。实验结果验证了提出算法的有效性,大大提高了复杂场景下的跟踪鲁棒性,有效地解决了部分遮挡和跟踪漂移问题。

Object tracking algorithm based on adaptive fusion of multi-feature

In view of partly occlusion and drifting problem in object tracking,we introduce an object tracking algorithm based on adaptive fusion of multi-feature and online learning.First,to improve the expression and discrimination of feature,we employ an improved feature extraction way.Then,we propose a new method,which uses adaptive fusion of multiple features with online learning of the detector and the object-specific recognizer.Finally,we employ challenging public testing dataset-PETS 2012 and use MOTP evaluate the performance of our method.Experiment results verify the effectiveness of our algorithm,greatly improving the tracking robustness in various environments.