PointpartNet++: Accuracy Improvement of 3D Point-Cloud Registration via Determination of Corresponding Points

This paper improves a deep learning model for point cloud registration of different sizes. Point is an important task applications such as 3D modeling, self-positioning mobile robots, and environment inspection. Recently, has started to deal with clouds. PointNet was the first classification semantic segmentation. Since then, methods based on tasks clouds have also been proposed. However, these are only suitable identical or nearly The size varies measurement distance, sensor type, environment, many other factors. Therefore, that need be registered may very sizes dimensions. Conventional cannot cope situations. In past, we proposed "PointpartNet", new neural network partial feature extraction. specialized in searching matching region between sizes, followed by simple method. this paper, improve increase accuracy robustness registration. done extracting local features passing them through explicit calculates correspondences registered. qualitative experiments, demonstrate vastly decreases error increases success rate s much 15 16 percent, compared previous research.