多目标跟踪中特征融合算子的选择技术研究

在基于信息融合的视频多目标跟踪过程中,特征融合算子的选择是关键.传统的基于特征融合的视频目标跟踪是将两个或两个以上特征作为测量特征通过某一融合机制进行融合权重.在一个高度非线性和难以建立模型的情形下,多目标之间的遮挡现象影响视频特性测量.为了解决该问题,提出了一种动态的选择融合算子的方法,以得到最好的跟踪性能.实验结果表明,动态地选择融合操作,可以提高互相遮挡的多目标跟踪性能.

Feature Fusion Operator Selection Technology Research of Multitarget Tracking

Abstract: Choosing feature fusion operator based on information fusion video multiple target tracking process is the key. The feature fusion methods of traditional video target tracking use two or more than two features as measuring features by a fusion mechanism. In the case of a highly nonlinear and hard to establish model, multiple targets scenario of repeated mutual occlusions influence video features measurement. In order to solve this problem, it is proposed a dynamic choice fusion operator method, to get the best tracking performance. The experimental results show that selecting dynamically fusion operator can improve the mutual occlused targets tracking performance.