RTFT6D: A Real-Time 6D Pose Estimation with Fusion Transformer

Accurate estimation of the 6D pose of objects is an essential prerequisite for robots to achieve precise object grasping. However, existing algorithms face challenges in accurately and robustly estimating the pose of target objects in robot visual grasping tasks. To confront the aforementioned challenges, we suggest a creative scheme that leverages RGB input and integrates transformer for 6D pose estimation of objects, obviating the need for any supplementary post-processing techniques. The network straightforwardly forecast the 2D projection of the 3D minimum bounding box of the target object, and then the PnP algorithm is utilized to recover the 6D pose of the object. Our method is tested on the LineMOD dataset, and its accuracy surpasses most RGB-input-based methods, and its speed is more than 10 times faster than almost all RGB-input-based methods—47 fps on Titan XP, with real-time pose estimation capability.