Normal-Based Flower Pollination Algorithm (FPA) for Solving 3D Point Set Registration via Rotation Optimization

Compared with the registration methods based on local optimizations, the heuristic registration methods are less sensitive to the initial position, and a reasonable bound range is essential to ensure the registration validity. In practice, compared with a rotation bound range, which is periodic, the setting of the translation range is more difficult and manual interventions required, especially when the initial position is complex. Moreover, it has yet to be discussed in past research. Therefore, a normal-based registration method based on the flower pollination algorithm is proposed in this paper, in which only rotation parameters (r
_x
, r
_y
, r
_z
) are considered. In our method, the point correspondences are guided by their normal due to their invariance to position translation. Considering the normal degeneration caused by noise, outliers, and partial overlapping, the Pauta criterion is employed to remove distorted correspondences and acquire reliable translation. Moreover, the population of optimal pollens is guaranteed by the use of the searching radius adjustment and periodic boundary. A number of experiments demonstrate that the proposed method exhibits competitive or better performance in terms of initial position, noise, outliers and partial overlapping. Furthermore, a real quality inspection is also implemented to confirm the availability and superiority of the proposed method in the manufacturing process.