Novel dynamic template matching of visual servoing for tethered space robot

Abstract

Aiming at the problems resulted from small template matching with large scene, such as low robust and long time-consumption, we propose a novel template matching algorithm. First, we define a similarity method called Normalized SAD, which is then followed by an improved matching criterion using grey and gradient values arranged into hollow annulus structure. Hollow structure can help decrease the accumulative deviation caused by the changes of background and the compound values make it restrain the impact of noises and illumination. Furthermore, we design a least square integrated predictor and updating strategy of dynamic template for robust tracking in each frame. Finally, the results of ground experiments indicate our algorithm can realize real-time recognition in complex circumstance.