TARGET POSITION ESTIMATION USING MULTI-VISION SYSTEM IMPLEMENTED ON DISTRIBUTED MOBILE ROBOTS

Abstract

In this study, a multi-vision system implemented on distributed mobile robots to estimate the target position of an object in the real world was developed. A triangulation method using the colour feature of an object was applied to develop the target position estimation method. The geometrical relationship among mobile robots affects the accuracy of target position estimation; thus, the effects induced by the geometrical relationship were analysed and a fusion method to obtain accurate estimation results was developed. Moreover, a video stabilization method that integrates the block matching method, weighs averaging method, and the Kalman filter design was developed to provide stabilized images to improve the target position estimation accuracy when the distributed mobile robots move on an uneven ground. A method for detecting and eliminating incorrect estimations was further developed to reduce the adverse effects induced by ill-conditioned image quality. Finally, an application related to the cooperative motions of multiple mobile robots demonstrated the feasibility of the multi-vision system and the approaches developed in this study using image- and position-visual servoing techniques.