Abstract
This thesis presents a line of study on finding good hand gestures for human-computer interaction. We aim to know why the frequently used hand gestures are effective and whether there are other good choices of hand gestures that might not be so obvious but actually are easy to recognize for the computers. Our approach is based on building a system for hand gesture classification. We generate a lot of training samples of synthesized hand gestures, and train a multiclass classifier to distinguish different gestures by their visual features. The gestures that are of high recognition accuracy can be considered as good candidates for the use in human-computer interaction. We also take into account the easiness of performing the gestures using a regression-based evaluation procedure. In the experimental results we show a ranking list of good gestures that are easy to recognize and easy to perform.