The Japanese Sign Language using general movement traits and localized characteristics in images

摘要

This thesis will report on the method of Japanese Sign Language recognition using general movement traits and localized characteristics in images captured by a stereo camera as well as test results of recognition experiment carried out on 4 test subjects. The sign language words used for recognition here are about 150 words expected to be used in a conversation at a post office counter. The hardware structure is as follows: two-eyed camera, image synthesizer for synthesizing of stereo images into a single image, color changing board for converting colors in input image, frame grabber for image capturing and personal computer for recognition processing.