A Robust CoS-PVNet Pose Estimation Network in Complex Scenarios

• DOI: 10.3390/electronics13112089
• Publication Date: 2024-05-28T07:48:49Z
• Authors: Jiu Yong; Xiaomei Lei; Jianwu Dang; Yangping Wang
• Keywords: Robustness

Object 6D pose estimation, as a key technology in applications such augmented reality (AR), virtual (VR), robotics, and autonomous driving, requires the prediction of 3D position objects robustly from complex scene images. However, environmental factors occlusion, noise, weak texture, lighting changes may affect accuracy robustness object estimation. We propose robust CoS-PVNet (complex scenarios pixel-wise voting network) estimation network for scenes. By adding pixel-weight layer based on PVNet network, more accurate pixel point vectors are selected, dilated convolution adaptive weighting strategies used to capture local global contextual information input feature map. At same time, perspective-n-point localization algorithm is accurately locate 2D points solve objects, then, transformation relationship matrix projection solved. The research results indicate that LineMod Occlusion datasets, has high can achieve stable even