SyMFM6D: Symmetry-aware Multi-directional Fusion for Multi-View 6D Object Pose Estimation

Detecting objects and estimating their 6D poses is essential for automated systems to interact safely with the environment. Most pose estimators, however, rely on a single camera frame suffer from occlusions ambiguities due object symmetries. We overcome this issue by presenting novel symmetry-aware multi-view estimator called SyMFM6D. Our approach efficiently fuses RGB-D frames multiple perspectives in deep multi-directional fusion network predicts predefined keypoints all scene simultaneously. Based an instance semantic segmentation, we compute least-squares fitting. To address ambiguity issues symmetric objects, propose training procedure keypoint detection including new objective function. SyMFM6D significantly outperforms state-of-the-art both single-view estimation. furthermore show effectiveness of our demonstrate that robust towards inaccurate calibration dynamic setups.