Tracking of infrared small-target based on improved Mean-Shift algorithm

Small-target tracking in infrared imagery with a complex background is always an important task in object tracking fields. Small and manoeuvrable objects in complex clutter and highly noised background usually results in serious false alarm in target tarking for low contrast of infrared imagery. An improved Mean-Shift algorithm to handle the influnce of complex background during tracking the smalltarget in infrared imagery was proposed. This work proposed an adaptive nonlinear machine to help Mean-Shift algorithm to get stable histogram of the interesting areas. This machine expanded the imformation of object histogram refer to the mean value of tracking window, as well as reduced the noise part of tracking window refer to the standard deviation of it. At the same time, algorithm fused the histograms of gradient with histogram of gray-value to discribe the target. To validate the effection of the proposed algorithm, the last part conduct a series tracking experiments which choose highly noised and clutered videos as their candidates. The comparison of the tracking results between tradtional Mean-Shift algotithm and improved Mean-Shift algorithm shows that the proposed algorithm has a more accurate tracking effection. Further more the proposed algorithm highly reduced the wobbleing between small-target and tracking window. This indicates that the improved algorithm achieved more robustness.