CNNを用いた物体の位置・姿勢推定

抄録

In this research, we propose a method for estimating 6 DOF object pose from a single RGB image based on convolutional neural networks (CNN). The estimated pose will be used as an initial position to run the Iterative Closest Point (ICP), which uses depth data to get the final position of the object. This approach is suitable for practical application of robot grasping an object. Unlike large scale database for object detection, the proposed system is trained with minimal datasets which can be obtained in a local environment. Users in different environment will be able to train the network suited for their own environment. The results show average error of 18.9 degrees, which we empirically found that it is low enough to successfully run the ICP algorithm.