Study Of Static Gesture Recognition On Radiant-projection-transform And Polar-radius-invariant Moment

As the development of human-computer interaction and pattern recognition, Human Gesture recognition is becoming one of the most important branches of these fields. It is widely studied in Human-Robot interactive. A large number of researches have approached some achievement. However, the Vision-based gesture recognize is weak on theory. The gestures' segmentation and recognizing are complex. Gesture classes for recognize remain scanty.In the dissertation, we provided a method to normalize gesture image. Using the method, the normalization image is invariant to scaling and transform. Then, we detect the orientation of the gestures automatically. With the gesture orientation, the wrist can be eliminated efficiently. Those works can improve the recognizing rate a lot.This dissertation gives a completely new method to do research on gesture recognition using radiant information. The Radiant-Projection-Transform method and Polar-Radius-Invariant Moment method are two image processing method based on image radiant information. Gesture in the image is a radiant region in which fingers radiant from palm. Different gesture shapes can be recognized by different fingers distributions' difference. In the gesture recognizing research, there are not such theories that using the radiant information recognizing gesture definitely.Radiant- projection- transform (RPT) is a new method for gesture recognition. Through a series of mathematic transform such as normalization and Fourier-transform, the gesture features have many excellent characteristics such as translation invariance, scaling invariance and rotation invariance. These features can be used to do study in pattern recognition. In the dissertation, we improve the RPT method. By changing the object gesture expressing, a better recognizing rate can be achieved. And by using it recognize edge gesture image, recognition speed can be improved a lot. We recognize static human gestures using this method in the dissertation. In the end, some experimentations show that it does good job in the gesture databases, and it is a way researching gesture recognition.In the dissertation, we also use the Polar-Radius-Invariant Moment method to recognize gestures. In the experiment, it is validated that PRI moment is a good implement to express gesture features and identifying them. We provided a series of features for gesture recognition. By recognizing gestures with edge images, we achieved a good result and the recognizing speed is improved.