3D Structure Estimation from a Single View Using Generic Fitted Primitives (GFP)

Abstract

This paper presents a method for surface estimation applied on single viewed objects. Its goal is to deliver reliable 3D scene information to service robotics application for appropriate grasp and manipulation actions. The core of the approach is to deform a predefined generic primitive such that it captures the local geometrical information which describes the imaged object model. The primitive modeling process is performed on 3D Regions of Interest (ROI) obtained by classifying the objects present in the scene. In order to speed up the process, the primitive points are divided into two categories: control and regular points. The control points are used to sculpt the initial primitive model based on the principle of active contours, or snakes, whereas the regular points are used to smooth the final representation of the object. In the end, a compact volume can be obtained by generating a 3D mesh based on the newly modified primitive point cloud. The obtainedPoint Distribution Models (PDM) are used for the purpose of precise object manipulation in service robotics applications.