L6DNet: Light 6 DoF Network for Robust and Precise Object Pose Estimation With Small Datasets

Estimating the 3D pose of an object is a challenging task that can be considered within augmented reality or robotic applications. In this letter, we propose novel approach to perform 6 DoF estimation from single RGB-D image. We adopt hybrid pipeline in two stages: data-driven and geometric respectively. The step consists classification CNN estimate 2D location image local patches, followed by regression trained predict set keypoints camera coordinate system. To extract information, aligning points system with corresponding world minimizing registration error, thus computing pose. Our experiments on standard dataset LineMod show our more robust accurate than state-of-the-art methods. also validated achieve positioning visual servoing.