Comparison of neural network and hausdorff distance methods in American, British and Turkish sign languages recognition

Abstract

This paper demonstrates a comparison between two approaches for developing a universal sign language recognition system. Both proposed approaches were verified for three different sign languages (American Sign Language, British Sign Language and Turkish Sign Language) and were implemented successfully. The first approach is designed and implemented by using feed forward neural network structure on Matlab while the second approach is mainly based on Hausdorff distance algorithm and it is implemented by using OpenCV libraries. Such systems are useful and convenient for hearing impaired people as they can easily communicate and no longer face difficulties in many aspects of their lives. Furthermore, the results of the first approach which was tested for all the three sign languages, achieved a 93.4% success rate whereas the second approach attained a less success rate with 90.9% percentage.