中圖分類號： TP391
文獻(xiàn)標(biāo)識碼： A
DOI： 10.19358/j.issn.2096-5133.2021.06.010
引用格式： 侯大偉. 一種基于實(shí)例分割和點(diǎn)云配準(zhǔn)的六維位姿估計(jì)方法[J].信息技術(shù)與網(wǎng)絡(luò)安全，2021，40(6)：56-61.

6D pose estimation based on instance segmentation and point cloud registration

Abstract： This paper proposes a method to estimate the six-dimensional pose of an object based on the Mask R-CNN instance segmentation network and the Super4PCS point cloud registration algorithm. The six-dimensional pose of the target can be generally estimated by registering the point cloud of the environment and the target. The number of point clouds generated by the scanning the overall environment with 3D equipment is huge, the calculation time will be too long when directly using the source point cloud to register the reference point cloud. Therefore, this article will divide the target before registration. Firstly, we obtain the RGB-D map of the overall environment by the depth-sensing camera. Secondly, we use the Mask-R CNN model to segment the target, after that we convert the RGB-D map of the segmented target into a point cloud image and complete the point cloud registration of the reference and the segmented target through Super4PCS algorithm. We finally obtain the six-dimensional pose of the target and compare the four sets of experiments based on our dataset. The results show that the time reduction rate is about 60%-80%, which effectively illustrates the feasibility of our method.

Key words : 6D pose estimation；Mask R-CNN instance segmentation；Super4PCS point cloud registration