Automatic Target Tracking in Infrared Image Sequences Using Ensemble Distance Metric

摘要：

This paper presents a novel algorithm named Ensemble Distance Metric Tracking (EDMT) for target tracking in infrared imagery. Obtaining an appropriate distance metric function can significantly improve the performance of tracking algorithms. There are two problems in distance metric choosing for object tracking. First, we cant find the data model distribution beforehand for most tracking application. Second, the data model will change as both foreground and background appearance undergoes complex changes with the target object moving from place to place. So the distance metric function also needs to adapt dynamically during the tracking procedure. Most tracking applications are conducted using a fixed distance metric function, which is determined beforehand. We propose a new algorithM that can learn and update the distance metric dynamically, which is different from the conventional methods that use the predefined metric. In our new EDMT algorithm, the ensemble distance metric function is learnt by weighted training with different distance metrics on each feature element iteratively using the boosting learning method. The new distance metric function is adopted in particle filter to compute the weights of each particle. The experimental results demonstrate the effectiveness and robustness of our tracking algorithm in challenging infrared video sequences.

关键词： Object tracking Ensemble learning Distance metric function Boosting

作者: Zhenyu Wang Guotian Yang

作者单位: National Engineering Laboratory for Biomass Power Generation Equipment, North China Electric Power U School of Control and Computer Engineering, North China Electric Power University, Beijing,102206, C

会议类型: 国际会议

会议名称: International Conference on Space Information Technology 2009(2009年第三届空间信息技术国际会议)

会议地点: 北京

会议语种:英文

页码: 1-7

在线出版日期: 2009-11-26（万方平台首次上网日期，不代表论文的发表时间）