3次元テンプレートマッチングによる指文字認識

抄録

This paper proposes a method for Japanese finger character recognition, using a 3-dimensional (3D) scanner. A hand is a complex dexterous manipulator, evolved to be more complex than any other animals. The hand, being capable of making many different complex shapes, it is ideal for communicating using gestures. The recognition of a whole language, such as the Japanese finger characters, requires the differentiation of subtle similar positioning of each digit. To know the exact 3D position of the hand's digits and overall shape, data gloves had been developed, but these are inconvenient to use. 2D image recognition systems struggle with recreating the 3D information. To capture the 3D information, the proposed method uses a 3D scanner, and then makes matches with 3D templates representing each unique character. Experimental results show that the proposed method recognizes a greater number of characters than existing 2D-based systems with recognition accuracy, on average of 93% for 9 testees, and a peak of over 98% for 4 of them.