PENGENALAN HAND GESTURE DINAMIS MENGGUNAKAN JARINGAN SYARAF TIRUAN MULTI-LAYER PERCEPTRON DENGAN METODE PEMBELAJARAN BACKPROPAGATION

The use of hand gestures to interact with a computer has multiple advantages over conventional input devices such as mouse or keyboard. Thus, it is necessary to develop a system that can recognize hand gestures and translated into a command to be executed by a computer. Many researches for recognizing hand gestures in real time still face challenges from hand position tracking problem and discoveries of optimum recognition method is needed. The developed system consists of acquisition, training and recognition process. Acquisition process is achieved using steps in digital image processing such as segmentation, dilation and erotion, convex hull and convexity defect to continuously estimate hand position and acquire hand movement path. Hand position coordinates is processed and normalized into image for training and identification. Training process uses backpropagation algorithm with hidden layer�s neuron count as varied parameters. Recognition result acquired from output of trained artificial neural network and interpreted into command for computer. In the experiments, the developed system has 96.97% success rate on hand gesture acquisition process if conducted at bright lighting room and between 50 and 60 cm from the webcam. Highest recognition rate acquired is 95.24% with 30x30 pixel normalized image size. Normalized image with 30x30 pixel size gives higher recognition rate than other size. Increasing training data from 330 to 550 improves average recognition rate from 78.04% to 88.96%