Recognition of sign language alphabet using colored gloves

This paper describes obtaining pictures of hands with a monocular camera, and building a system for recognizing characters of the manual alphabet for the deaf from the obtained image features. Considering that a handlike joint structure is capable of variegated deformation with many degrees of freedom, various problems are caused during shape recognition, such as finding equivalents, dealing with occlusions, and so forth. Therefore, in the current study, the authors proposed a method based on wearing colored gloves while taking pictures, which allowed easy detection of specific hand regions with improved accuracy, and handling of occlusions. In addition, the authors discuss the decision tree learning results obtained using a CAD ideal model which was computer-generated from the real image data, and also describe verification of the learning results based on real images.