TY - GEN
T1 - A Fast Coherent Point Drift Method for Rigid 3D Point Cloud Registration
AU - Liu, Zhengmao
AU - Yu, Chengpu
AU - Qiu, Fanshuo
AU - Liu, Yixuan
N1 - Publisher Copyright:
© 2023 Technical Committee on Control Theory, Chinese Association of Automation.
PY - 2023
Y1 - 2023
N2 - In laser dense mapping, registering large 3D point clouds can be a challenging task. The Coherent Point Drift (CPD) algorithm has been proven to be a superior method for point cloud registration in terms of accuracy. However, for large-scale point cloud data, the slow registration speed of CPD becomes a bottleneck. In this paper, a fast rigid registration method for 3D point clouds is proposed. The proposed method first models the point clouds as Gaussian mixture models and then uses EM algorithm to solve the transformation. Furthermore, based on improved fast gauss transform(IFGT), the proposed method introduces a tree data structure to search for the adjacent clusters of the target point and forms four methods to compute correspondence matrix, which is pretty time-consuming to compute in orginal CPD. The proposed method automatically selects the most efficient method among them. Finally, the optimal rigid transformation parameters are solved using the correspondence matrix posterior probability. Experimental results show that the proposed algorithm can speed up the registration while maintaining the same level of accuracy as the original CPD algorithm.
AB - In laser dense mapping, registering large 3D point clouds can be a challenging task. The Coherent Point Drift (CPD) algorithm has been proven to be a superior method for point cloud registration in terms of accuracy. However, for large-scale point cloud data, the slow registration speed of CPD becomes a bottleneck. In this paper, a fast rigid registration method for 3D point clouds is proposed. The proposed method first models the point clouds as Gaussian mixture models and then uses EM algorithm to solve the transformation. Furthermore, based on improved fast gauss transform(IFGT), the proposed method introduces a tree data structure to search for the adjacent clusters of the target point and forms four methods to compute correspondence matrix, which is pretty time-consuming to compute in orginal CPD. The proposed method automatically selects the most efficient method among them. Finally, the optimal rigid transformation parameters are solved using the correspondence matrix posterior probability. Experimental results show that the proposed algorithm can speed up the registration while maintaining the same level of accuracy as the original CPD algorithm.
KW - Coherent Point Drift (CPD)
KW - EM algorithm
KW - Gauss Mixture Model (GMM)
KW - Improved Fast Gauss Transform (IFGT)
KW - Point Cloud Registration
UR - http://www.scopus.com/inward/record.url?scp=85175529495&partnerID=8YFLogxK
U2 - 10.23919/CCC58697.2023.10240967
DO - 10.23919/CCC58697.2023.10240967
M3 - Conference contribution
AN - SCOPUS:85175529495
T3 - Chinese Control Conference, CCC
SP - 7776
EP - 7781
BT - 2023 42nd Chinese Control Conference, CCC 2023
PB - IEEE Computer Society
T2 - 42nd Chinese Control Conference, CCC 2023
Y2 - 24 July 2023 through 26 July 2023
ER -