Mean-shift tracking for IR characteristics of target based on online feature fusion

An improved mean shift tracking algorithm for IR target was proposed. Firstly, the local gray mean feature and local standard deviation feature were utilized to realize target modeling based on the low SNR characteristic of IR images. Secondly, according to the low contrast feature of target, the new feature-representing model was established where the feature likelihood ratios of target and local background were regarded as the weight value of kernel histogram.The final target representation model was obtained by means of linear fusing the two feature models, and the fusion coefficient was determined adaptively by contrast ratio of feature likelihood map. And lastly, the expression of shift vector in the process of the model gradient matching was derived in the framework of mean shift. Meanwhile, the discrimination criterion of model updating based on inter-frame change of the comprehensive contrast under complex background was constructed. The validity and the feasibility of the algorithm are proved by the actual experiments of IR target tracking.