Thai sign language translation system using upright speed-up robust feature and dynamic time warping

Sign language is an important communication tool for the deaf. There are two types of Thai sign language, i.e., Thai finger spelling (a signer spells each character using their fingers) and Thai sign language (a signer uses hand gesture and face expression to represent each word). In this paper, we build a dynamic hand gesture translation system with video caption without prior hand region segmentation. In particular, we utilize the upright speed-up robust feature (U-SURF) and the dynamic time warping (DTW) to find a matched word. This Thai sign language translation system was tested on 42 words. The total number of video sequences used in the experiment is 1470. The best correct classification rate for blind test set of signer-dependent is approximately 98% whereas that of signer-semi-independent is around 76 to 79%. The best blind test result for the signer-independent is around 62 to 68%.