SIFT-Cloud-Model generation method for 6D Pose estimation and its evaluation

A function to find objects and to estimate their 6D poses(x, y, z, pitch, yaw, roll) is crucial for a home service robot that works in human living environment. If a robot obtains 6D poses of target object, it can bring to that and use as tools. We have proposed SIFT-Cloud-Model (SCM) [1], that is designed to represent 3D objects with textures by 1) SIFT feature descriptors, 2) their 3D positions, and 3) their observed directions (hereafter "view vectors"). This model is used to find target objects in a scene and to estimate their relative 6D poses. In this paper, we propose the SIFT-Cloud-Model generation method based on structure from motion technique combined with optimization technique. Before this research, we had only one model. So we built this system of efficiently generating models to save our time and evaluated its to confirm the effectiveness of SCM to more objects. Finally experimental results of its accuracy evaluation and computational cost will be shown.