Rigid 3D Point Cloud Registration Based on Point Feature Histograms

Abstract

Depending on the displacement and orientation between point clouds, the registration of scattered point clouds is offten divided into two steps: crude and fine alignment. An approach of point cloud classification based on point feature histogram was proposed in this paper. We propose a method of establishing the point feature histograms to match feature points in different clouds. To reject the outliers, Random Sample Consensus algorithm is used. The rigid transformation matrix in crude alignment is then computed by Singular Value Decomposition method. The golden standard for fine alignment is the Iterative Closest Point algorithm and its variants. In this paper we apply a dynamic constraint of distance to improve the traditional algorithm. The experiment shows that our process of registration works fine with higher accuracy and efficiency.

Automatic point cloud registration algorithm based on the feature histogram of local surface

“…Prakhya S M calculated the HoNo (Histogram of Normal Orientations) at every point and detected the key point by evaluating the properties of both the HoNo and the neighborhood covariance matrix [15]. The point feature histogram (PFH) algorithm and the fast point feature histogram (FPFH) algorithm are popular algorithms of feature description [16][17][18], which generate a feature histogram for each point based on feature information. Prakhya S M et al applied a binary quantization method on a state-of-the-art 3D feature descriptor [19], SHOT [20], and created a new binary 3D feature descriptor, B-SHOT.…”