We propose a finger character recognition method using Microsoft Kinect. In our proposed method, we automatically extract the right hand region from depth image which captured by Kinect. Next, we calculate six features of number of finger, aspect ratio, area ratio between bounding rectangle and hand region, roundness, range ratio, and Foureir descriptors. These features are fed to SVMs. In order to prevent wrong recognition in a real-time process, we implement two functions of motionless judge process and voting process. We set 41 Japanese finger characters without a motion as the recognition target, and the evaluation experiments were carried out with five subjects. As the results, we obtained the recognition rate of 95% of person dependent recognition, and 53% of person independent recognition.%本論文ではMicrosoft社製Kinectを用いた指文字認識手法を提案する.Kinectより得られた距離画像から右手領域を自動的に抽出する.次に手領域より指の本数,アスペクト比,外接矩形と手領域の面積比,円形度,距離分布比およびフーリエ記述子の6種類の特徴量を求め,最後にSVMを用いて認識する.リアルタイム処理では,誤認識を防ぐために静止判定および投票処理を導入する.動きを伴わない41種類の日本指文字を認識対象とし,被験者5名に対して評価実験を実施した.その結果,特定人物認識実験では平均認識率95%,不特定人物認識では平均認識率53%の認識率を得た.
展开▼