The parameters effect on performance in ANN for hand gesture recognition system

The effectiveness of a neural network function depends on the network architecture and parameters. For discussing the relationship of parameters and performance, this study proposes a novel hand gesture recognition system (HGRS) combining the VICON and the back propagation neural network (BPNN). In this study, different numbers of hidden layer neurons and different numbers of layers were compared for effects on system performance. Too many or too few neurons reduced the recognition rate. Further, the hidden layer was needed for improving the system performance of the system. The training epoch size affects the general ability of the system. If the epoch size is too large, the system “over fit” the training set, and its general ability is impaired. However, an overly small epoch size would impair system recognition. The learning rate and system momentum affect the RMSE of the trained system. A higher learning rate and reduced momentum decrease RMSE.