抄録
We propose a method to genarate sight feature space for motion acquisition of robots The sight feature space is generated by embedding a vector composed of the points of features of the image to a low-dimensional space using LLE(Locally Linear Embedding). LLE is a manifold learning method for distribution of high-dimensional data to a low-dimensional space using its local linear structure. Image feature is extracted from the image including the arm of robot and the object using a SIFT(Scale Invariant Feature Transform).
