Web proceedings papers

Authors

Abstract

Albania Sign Language (AlbSL) is relatively new and little work has been done on sign recognition by using natural user interfaces (NUI) technologies. The paper aims to present an automatic, real-time solution for recognition of a limited set of Albanian sign language for the numbers from one to five by using Microsoft Kinect. Kinect device offers a depth sensor which is used to capture input data. Hand segmentation is performed on input data by applying a threshold constant. A contour tracing algorithm is used to extract the hand contour. Fourier descriptors based on centroid distance signature are generated and only first 15 Fourier descriptors coeficients are used to represent the hand shape. The resultant Fourier descriptors are then compared to each element in a training data set by using Euclidian distance. The element with the minimum distance is classified as a match. The proposed system achieved an accuracy of 96% and is able to process images with an average rate of 90 frames per second.

Keywords

AlbanianSignLanguage(AlbSL), FourierDescriptors, Number Recognition, Microsoft Kinect