複数視点からの観察による人姿勢認識のための視点計画

抄録

In this paper, we proposed viewpoint planning method for human pose recognition who lying inside building. Taking multiple viewpoints is efficient to recognize complicated object like human poses. However, the minimum viewpoint sift is ideal. So, our method adopted already proposed framework witch choose efficient viewpoint based on probability calculation. The effectiveness of this framework has been demonstrated in several daily object. We made changes to this demonstration by using human pose features extracted by convolutional neural network. This features obtain from middle layer the one before output layer of network have learned human posture. We evaluate this extracted features by compare with FPFH.