Recognition of Malaysian Sign Language using skeleton data with Neural Network

This paper proposes the Backpropagation Neural Network (BNN) for recognition of Dynamic Malaysian Sign Language. In the first stage, image acquisition data is obtained from the Kinect sensor, using skeletal data tracking with eight joint positions. In the second stage, for the skeletal feature extraction, the value of coordinate X, Y, Z of the data relative to the spine and head are taken. For the spherical coordinate conversion process, segmentation of the frame is done to get the same number of dimensions. Finally in the final stage, the classification process is performed using Backpropagation Neural Network (BNN), by variations of nodes in hidden layer. Experiments indicated that our system was able to recognize 15 dynamic Malaysian Sign Language with 80.54% accuracy.