6D Pose Estimating Using Multimodal Input and Attention Mechanism

6D pose estimation is a key problem in computer vision-based robotic manipulation grasping tasks. It is a challenging task due to the diverse lighting conditions, cluttered backgrounds, shadows, and occlusions in real scenes. To address these challenges, we propose a 6D pose estimation method that combines multimodal input and channel attention mechanism from monocular RGB images. Our method consists of two modules: The prediction module uses ResNet-18 as the backbone network, and adds a channel attention mechanism to estimate multiple intermediate representations of the 6D poses of objects, including keypoints, edge vectors, and pixel-wise symmetrical correspondence; The pose regression module uses the EPnP algorithm and SVD to regress the 6D poses of objects from the intermediate representation results. In the experiments, our method is evaluated on two public datasets, and the results show that our method can accurately estimate the 6D poses of the objects, and outperform the existing advanced methods under the ADD(-S) indicator, especially having good robustness in occluded environments.