基于联合直方图的红外与可见光目标融合跟踪

Tracking Infrared-visible Target with Joint Histogram

针对传统单核跟踪算法只能单独跟踪红外或可见光运动目标,导致目标的跟踪效果不是很理想,甚至跟踪失败的问题,本文提出了一种基于均值漂移的红外与可见光目标融合跟踪算法。该算法仍以直方图为目标表示模型,通过将红外目标的相似度和可见光目标的相似度进行加权融合,来构建新的目标函数,并依据核跟踪推理机制导出目标的联动位移公式;最后使用均值漂移程序实现目标的自动搜索。多个视频序列对的测试结果表明,本文提出的融合跟踪方法在处理场景拥簇、光照变化等方面要优于传统的单源跟踪方法,同时具有较高的实时性。

Due to traditional kernel tracking algorithm can only track infrared or visible target,its performance is poor,even unsuccessful.This paper proposes a fusion tracking method for infraredvisible target by using a mean shift algorithm.Firstly,the histogram is still adopted to represent the infrared target and visible target,and the similarity between infrared candidate and its target,and the similarity between visible candidate and its target,are integrated into a novel objective function with different weight.Secondly,similar to mean shift on the objective function,ajoint target location-shift formula is induced to the new method.Finally,the optimal target location is gained recursively by applying the mean shift procedure.Experimental results of several infrared-visible image sequences demonstrate that the proposed fusion algorithm is superior to the single-sensor tracking algorithm in handling illumination change and background clutter.