Efficient 3D object recognition via geometric information preservation

作者：

Highlights：

• A new unsupervised encoder layer is able to extract 3D point-wise features directly.

• Improving feature discrimination performance via geometric information preservation.

• Improving robustness by stacking locally aggregated features on point-wise features.

• Using 3D point cloud features better resists occlusion and background clutter.

摘要

•A new unsupervised encoder layer is able to extract 3D point-wise features directly.•Improving feature discrimination performance via geometric information preservation.•Improving robustness by stacking locally aggregated features on point-wise features.•Using 3D point cloud features better resists occlusion and background clutter.

论文关键词：Stacked 3D feature encoder,3D object recognition,6-DOF pose estimation,Geometric information preservation

论文评审过程：Received 8 October 2018, Revised 25 February 2019, Accepted 23 March 2019, Available online 26 March 2019, Version of Record 29 March 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.03.025