A Preliminary Study on 3D Model Reconstruction Using Motion Vectors Contained in Motion Picture

Abstract

The 3D point cloud is a field of computer vision，and the demand for it has been increasing in recent years due to the spread of 3D surveying instruments．Since point cloud data shows the shape of an object，it is expected to be applied in various fields such as robots，architecture / civil engineering，transportation，and agriculture．In this research，we focused on the technology for reconstructing 3D point cloud data from moving image series among the SfM (Structure from Motion) technologies for reconstructing 3D point cloud data from images．When reconstructing a 3D model from an image，the processing time and accuracy of the reconstruction differ depending on the image used，and it is necessary to select an appropriate image for efficient reconstruction．However，it is difficult to manually select an appropriate image from many image groups．In this study，we investigated a method of reconstructing a three-dimensional point cloud by selecting a frame using the motion vector contained in motion picture．The experimental results using some moving image data are shown and given consideration.