Feature Description using Planar Surfaces for 3D Virtual Keypoint

Abstract

In recent years, the shape analysis of the 3D point cloud has become an important research theme of computer vision.This process is expected applying to the modeling, the object recognition and the localization, however mainly requires huge computation time for amount of data of point cloud. In order to solve this issue, the approach to extract a small number of feature points having repeatability and distinctiveness is effective. On the other hand, the performance of conventional feature point is strongly dependent on the sensor noise and resolution, because its extraction method selects actual points.We proposed a novel method to allocate the keypoints from multiple plane equations to a virtual location without selecting from actual points. In this report, we proposed a novel feature description method that added rotation invariant. Moreover, we confirmed that proposed method is superior than conventional method by performance evaluation of keypoint detection and feature description.