Coarse registration of dense point clouds based on image feature points

Abstract

Different view point clouds of objects can be achieved through machine vision and need to be translated into a coherent coordinate system for registration. To reduce the number of iterations of accurate registration algorithm and avoid local optical algorithm, coarse registration can provide the initial value of good posture for precise registration. For solving the issues, we proposed a practical coarse registration method of point clouds based on image feature points. In the proposed method, 3D feature point detection methods were respectively established based on dense point clouds derived from monocular structured light vision according to 2D feature points detected by using Speeded Up Robust Features (SURF) algorithm. Through the combination of rigid body posture measurement method and removing method of gross error points, the precise rotation matrix and translation vector before and after moving point clouds were obtained. In the experiment, we introduced the method of dense point clouds derived from structured light in detail. The experimental results indicated that the method could provide good initial postures for precise registration of point clouds.