In robotics, realizing 6D tracking, that is, tracking the three-dimensional position and orientation of moving objects in real-time, is a significant challenge. In this study, we extend fast pixel-wise posterior 3D (Fast-PWP3D), a method for estimating the position and posture simultaneously. The extended method involves the segmentation of the target region from multiple camera images using a 3D target model. The energy function was modified to handle multiviewpoint posture estimation. Consequently, the accuracy of target estimation and robustness against occlusion were improved.
Mizuno et al. (Tue,) studied this question.