| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 89 === Abstract
In the recent years, due to the popularity of personal computers, people try to find a more natural way to communicate with the computer beyond the keyboards and the mouse enthusiastically. In addition to the speech recognition systems, hand gesture has become a new man-machine interface tool.
In this research we use the HMM-based hand gesture recognition system to recognize 10 different gestures in a simple background. We test three different kind of features, namely, the Fourier descriptors, the energy spectrum and the sum of energy spectrum, as the input to the HMM to discriminate the contour of hand shapes. The whole system involves feature extraction, training and recognition phases.
In the feature extraction phase, we get the symbol sequence from the associated input gesture sequence by means of the preprocessing and vector quantization. In the training phase, we use 20 different training samples for each gesture to build the HMM by using the Baum-Welch algorithm. In the recognition phase, while a symbol sequence obtained from an input gesture sequence is given, we use the forward-procedure to compute the probability of each model producing the symbol sequence, and select the model with the maximum probability as the recognition output.
Based on the experiment results, the recognition rate by using the Fourier descriptors, the energy spectrum and the sum of the energy spectrum are 94%, 98% and 97.7% respectively. This shows that the HMM-Based gesture recognition system works well especially when the energy spectrum proposed in this research is used.
|
|---|
