Neural Network Performance Analysis for Real Time Hand Gesture Tracking Based on Hu Moment and Hybrid Features

This paper presents a comparison study between the multilayer perceptron (MLP) and radial basis function (RBF) neural networks with supervised learning and back propagation algorithm to track hand gestures. Both networks have two output classes which are hand and face. Skin is detected by a regional based algorithm in the image, and then networks are applied on video sequences frame by frame in different background (simple and complex) with different illumination of environment to detect face, hand and its gesture. The number of training and testing samples in networks are equal and the set of binary images obtained from skin detection method is used to train the networks. Hand gestures are 6 cases which are tracked and they were not recognized. Both left and right hands has been trained to the network. Network features are based on the image transforms and they should not relate to deformation, size and rotation of hand. Since some of the features are in common with each other so a new method is applied to reduced calculation of input vector. Result shows that MLP has high accuracy and higher speed in tracking hand gesture in different background with minimum average error but it has a lower speed in training and convergence compare to the RBF in its final average error.