Improving Inter-Person American Sign Language Recognition Accuracy Using Deep Learning

Chang, Pei Woon (2019) Improving Inter-Person American Sign Language Recognition Accuracy Using Deep Learning. Project Report. UTeM, Melaka. (Submitted)

Text (24 Pages) Improving inter-person american sign language recognition accuracy using deep learning.pdf - Submitted Version Download (319kB)

Abstract

Most of the normal people don’t understand sign languages performed by deaf-mute people, thus a sign language to text/voice conversion systems would greatly improve the communications between them. Hence, deep learning-based sign language system is proposed to recognize the muscle signal into a different type of American Sign Language. Deep Feed-Forward Neural Network (DFFN) and Shallow Convolutional Neural Network (SCNN) are used for this study. The proposed model evaluated with collected datasets from six subjects. A Myo Armband is used to collect the sEMG from right forearm for ten different sign language. Two experiments are conducted to evaluate the performance of both networks in intra-person and inter-person. SCNN has higher accuracy (99.7%) than DFFN (87.93%) in intra-person. However, there is no significant difference in both networks in the inter-person. Both the DFFN and SCNN show an improvement in the accuracy for inter-person by 5.82% and 5.62% respectively when more subjects are included in the training datasets. Based on the result, the inter-person accuracy can be improved by using a generalized dataset.

Actions (login required)

View Item