@techreport{oai:ipsj.ixsq.nii.ac.jp:00176649, author = {小島, 聖司 and 大山, 航 and 若林, 哲史 and Seiji, Kojima and Wataru, Ohyama and Tetsushi, Wakabayashi}, issue = {52}, month = {Jan}, note = {本研究では調理動作認識手法を提案する.本研究の問題設定は以下の通りである.(1) 事前に定められた複数のメニューのうちのひとつを調理する.(2) 複数の調理者がそれぞれのメニューを調理する様子を撮影した映像が,学習用映像として与えられる.(3) 学習用映像の各フレームには調理動作のラベルが付与されている.(4) 提案手法は学習用画像を用いて識別器を学習し,入力された調理映像の各フレームに調理動作のラベルを付与する.動作認識手法には,HOG 特徴を時間方向に拡張した時空間 HOG 特徴を利用して,SVM による認識を行った.Kitchen Scene Context based Gesture Recognition (KSCGR) データセットを用いて性能を評価した.評価の結果,認識成功率 76.5%,KSCGR 評価スコア 71.5% となった., In this research, we propose a cooking gesture recognition method. The outline of this research is as follows, (1) Actors cook one of preliminary determined menus. (2) Sequential gestures of their cooking are captured as image sequences, and these sequences are given as a training data. (3) Each frame of training data is annotated by a cooking gesture label. (4) In the proposed method, classifier are trained using frame images in training data, and labels of cooking gesture are assigned to each frame of the input cooking images. In the proposed method, the test data is recognized by SVM using spatio-temporal HOG featrues which extended HOG features in the time domein. We evaluated the performance using Kitchen Scene Context based Gesture Recognition (KSCGR) data set. As a result of the evaluation, the recognition accuracy of 76.5% is obtained and the KSCGR evaluation score achieves 71.5%.}, title = {時空間HOG特徴を用いた調理動作認識\n}, year = {2017} }