多特征联合匹配的目标图像稳定跟踪算法

Stabilized Target Tracking Algorithm Using Multi-feature Joint Matching

针对图像制导目标跟踪系统中应用单一特征跟踪算法,由于特征不稳定而造成的跟踪点漂移,甚至目标丢失等问题,提出多特征联合匹配的目标图像复合跟踪方法。分析了分别单独使用灰度特征或者点特征进行目标跟踪存在的问题;通过归纳常用的目标跟踪算法与特征类型的对应关系,依据经验给出了几种多特征组合方案;以灰度特征和点特征作为多特征联合匹配依据进行方法验证。算法利用两种特征分别获取跟踪点,依据跟踪点邻域与模板之间的相似性度量准则,确定两种方法获得的跟踪点与理想目标点的一致程度,并为其分配权值,联合确定最佳跟踪点,并作为模板更新依据。以某导弹导引头实际飞行视频数据,进行多特征联合匹配跟踪点仿真。通过对比单一方法,算法综合了多种特征的优势,且跟踪点漂移范围满足要求。对进一步改进跟踪轨迹平滑性,以及算法实时性实现进行了探讨。

The single feature tracking algorithm of imaging guidance system has a disadvantage of feature instability which can cause the shift of tracking point and the loss of targets in the course of tracking.A multi-feature joint matching method for target tracking was proposed.The drawbacks of tracking algorithm using gray or point feature were analyzed,the corresponding relationship between target tracking algorithm and feature type was induced,and several multi-feature modes were given according to the experiences.The validity of the method using gray and point features as multi-feature was demonstrated.The algorithm is use the gray feature and point feature of target image to calculate the tracking points,respectively.The consistency between the tracking point of each tracking algorithm and the ideal target points was determined according to the measure criteria of comparability between tracking point neighbor and template,a weight coefficient was allocated for each matching algorithm,and an optimal tracking point was given.Besides,the tracking point was taken as a basis of template update.The algorithm overcomes the shortcoming of single tracking method and realizes the stabilized target tracking under the complex background.At last,the tracking video data of a certain missile was used to demonstrate the validity of multi-feature tracking algorithm,compared with a single feature method,the algorithm combines the virtues of multiple features,and the calculated tracking point shift is small.And also,the further improvement of tracking trajectory smoothness and the real-time implementation of algorithm were investigated.