International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD2023)

125 / 2023-09-20 15:39:57

Adaptive Weighted 6d Pose Estimation Method using RGB-D Data

deep learning,RGB – D,6D pose estimation,robotic grasping

Special Sessions > SS08 - IoT, AI Methods, and Applications in Health Monitoring and Diagnosis of Structure, Power, Medicine, etc

Aiming to address the challenge of underutilizing color and depth information when estimating the six-degree-of-freedom (6-DoF) target pose for a single RGB-D image, a deep learning framework based on pixel-wise adaptive feature embedding fusion network is proposed. This method is employed for estimating the 6-DoF poses of a known set of objects in highly cluttered scenes. The network is tested on the publicly available LineMOD dataset, and experimental results demonstrate that, compared to existing methods for 6-DoF pose estimation of similar types, the model presented in this paper achieves superior accuracy in predicting the 6-DoF poses. When evaluated using the same criteria, the average accuracy reaches 96.5%.