ICP Algorithm

ICP Algorithm in 2D and 3D

1) Introduction

For 3D registration of point sets of an object, we need point sets from at least two different points of view. And the merging process is needed. A popular approach to this process is Iterative Closest Point (ICP) algorithm introduced by Besl [1]. I improved basic ICP algorithm by T. Suguihara's LM method [4] based on A. W. Fitzgibbon's paper [2]. And the results of my ICP algorithm is in section 2. Now, I continuously try to improve this algorithm by searching others paper and from my idea.

2) Results

A. 2D-ICP algorithm

Video 1. 2D ICP algorithm

Fig 1. Before processing

Fig 2. After processing

B. 3D-ICP algorithm

Video 2. 3D ICP algorithm

Fig 3. Before processing

Fig 4. After processing

Discussion of the detailed process and results is in my report [5].

(All users can view my Report by click the reference [5].)

3) References

[1] P. J. Besl, N. McKay, A method for registration od 3-D shapes, IEEE Transactions on Pattern Analysis and Machine Intelligence 14 (2) (1992) 239 - 256.

[2] A. W. Fitzgibbon, Robust registration of 2D and 3D point sets, Image Vision Computing., vol. 21, no. 13–14, pp. 1145–1153, 2003.

[3] S. K. Chan and P. D. Lawrence, "General inverse kinematics with the error damped pseudoinverse," Proceedings. 1988 IEEE International Conference on Robotics and Automation, pp. 834-839 vol.2, 1988.

[4] T. Sugihara, "Solvability-Unconcerned Inverse Kinematics by the Levenberg–Marquardt Method," in IEEE Transactions on Robotics, vol. 27, no. 5, pp. 984-991, Oct. 2011.

[5] Jae Hoon An, "ICP algorithm", Jun. 2021.

[6] 3D data : "Stanford Bunny" @ Stanford University Computer Graphics Laboratory