Cognitively Inspired 6D Motion Estimation of a Noncooperative Target Using Monocular RGB-D Images

Abstract

With the increasing demand of information acquisition in cognitive computation, there has been a growing interest in the study of cognitively inspired computation methods for motion estimation through the use of the vision to obtain more information. Generally, when implementing those methods, it is, however, necessary on occasion to know some prior information of the target complying with some rules. It therefore imposes some obstacles to the application of those approaches. In this paper, a novel cognitively inspired 6D motion estimation method for a noncooperative target is proposed through the use of monocular RGB-D images. We build the 3D model of target first and then estimate the pose of target by solving the perspective-n-point problem. In addition, the Kalman filter is applied to estimate the 6D motion of a target. The target estimated in our method can move with arbitrarily varying velocity on the practically reasonable assumption that those acquired images are not distorted. In our estimation method, priori information of the target and the constraint of target moving velocity are not needed. Simulation and experimental results are provided to demonstrate the effectiveness of our method.