INDOOR SCENE REGISTRATION BASED ON SIAMESE NETWORK AND POINTNET

Abstract. This paper presents a deep learning feature-based method for registration of indoor mobile LiDAR data. Our method is to input point cloud directly, which is more robust to noise than traditional algorithms. The proposed method involves three steps. We first extract the key points by Harris3D algorithm and get their local patches by our sampling method. Second, a Siamese network is trained to describe the patches as local descriptors. Finally, we obtain the final matching pairs depends on the distance which is between two descriptors, and then solve the transformation matrix. The accuracy of registration is within 6 cm when the overlap is greater than 35%. In order to improve the registration accuracy, the ICP algorithm is used to fine-tuning the registration results. And the final registration accuracy is within 3.5 cm. The experiments show that our method applied to the registration of indoor mobile LiDAR data robustly and accurately.

[1]  Chenglu Wen,et al.  A Target-Free Automatic Self-Calibration Approach for Multibeam Laser Scanners , 2018, IEEE Transactions on Instrumentation and Measurement.

[2]  Cewu Lu,et al.  PointSIFT: A SIFT-like Network Module for 3D Point Cloud Semantic Segmentation , 2018, ArXiv.

[3]  Matthias Nießner,et al.  3DMatch: Learning Local Geometric Descriptors from RGB-D Reconstructions , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[4]  Andrew E. Johnson,et al.  Using Spin Images for Efficient Object Recognition in Cluttered 3D Scenes , 1999, IEEE Trans. Pattern Anal. Mach. Intell..

[5]  Benjamin Bustos,et al.  Harris 3D: a robust extension of the Harris operator for interest point detection on 3D meshes , 2011, The Visual Computer.

[6]  Mohammed Bennamoun,et al.  Rotational Projection Statistics for 3D Local Surface Description and Object Recognition , 2013, International Journal of Computer Vision.

[7]  Federico Tombari,et al.  Unique Signatures of Histograms for Local Surface Description , 2010, ECCV.

[8]  Radu Bogdan Rusu,et al.  3D is here: Point Cloud Library (PCL) , 2011, 2011 IEEE International Conference on Robotics and Automation.

[9]  Xiao Xiang Zhu,et al.  A CNN for the identification of corresponding patches in SAR and optical imagery of urban scenes , 2017, 2017 Joint Urban Remote Sensing Event (JURSE).

[10]  Andrew Zisserman,et al.  Descriptor Learning Using Convex Optimisation , 2012, ECCV.

[11]  Jonathan Li,et al.  Pairwise registration of TLS point clouds using covariance descriptors and a non-cooperative game , 2017 .

[12]  Gernot Riegler,et al.  OctNet: Learning Deep 3D Representations at High Resolutions , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[13]  Leonidas J. Guibas,et al.  PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[14]  Leonidas J. Guibas,et al.  ShapeNet: An Information-Rich 3D Model Repository , 2015, ArXiv.

[15]  Leonidas J. Guibas,et al.  PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space , 2017, NIPS.