로봇 물체 조작을 위한 딥러닝 시맨틱 분할을 이용한 물체 검출

Object detection is the first necessary step for the manipulation of objects with robot. Object detection is used to find the type and number of a specific object on an image and identify the place where the object is located. In this paper, we adopt semantic segmentation method for the robust detection of objects in an image. We use the FCN[6] architecture which consists of an encoder and decoder for the semantic segmentation. Transfer learning is adopted where the parameters of the encoder part are used as the one which are learned by ImageNet and the parameters of the decoder part are newly learned by using training images of given tasks. We deal with three classes including two objects and one background for the semantic segmentation. The experimental results show the feasibility of the presented algorithm.