1P1-E08 SIFT-Cloud-Modelを用いた物体の発見及び6次元姿勢推定とGPGPUによる高速化

概要

論文の詳細を見る
A function to find objects and to estimate their poses is crucial for a home service robot that works in human living environment. In this paper, we propose a "SIFT-Cloud-Model (SCM)", that is designed to represent objects in daily scene. The model consists of SIFT feature descriptors, and their 3D positions, and eye vectors. The model is automatically obtained by using Structure from Motion techniques. Then it is matched against single camera images by finding the corresponding SIFT features, arid using them to estimate the object pose. The system is optimized using GPGPU methods. Finnally experimental results for accuracy and calculation time are shown. The method achieves 5.5 fps with 8% error in translation and 25deg. error in rotation.
一般社団法人日本機械学会の論文