Font Size: a A A

Hand Feature Extraction Using The Curvature Scale Space Descriptor

Posted on:2016-06-03Degree:MasterType:Thesis
Country:ChinaCandidate:C LiFull Text:PDF
GTID:2308330503950640Subject:Computer technology
Abstract/Summary:PDF Full Text Request
One of an important way of new human-computer interaction research is gesture interaction. This interaction is natural and non-contact interaction. So in the future the trend of the development of the human-computer interaction will be more tend to gesture interaction. Gesture recognition technology involved in computer graphics, pattern recognition, artificial intelligence, and many other subjects. Therefore, the study of gesture recognition have very important research value and research significance.The current research based on the hand articulation detection focuses on the optical image processing, including hand detection, gesture recognition and hand feature points extraction. Hand detection is used for find hand, mainly divided into static gestures and dynamic hand gestures. Gesture recognition is mainly used for generated the results to identify the feature points of the target. Hand feature extraction is divided into two kinds of method, that are hand feature extraction based on hand contour and hand feature extraction based on hand model. Therefore, in order to achieve efficient and accurate non-contact human-hand characteristic points recognition, this paper proposes a human-hand characteristic points recognition method based on curvature scale space, with front hand gestures as the research object, put forward a kind of non-contact detection staff characteristic points recognition methods, our work is as follows:(1)we introduce two kinds of method of traditional hand feature extraction which is based on hand contour, that are convex polygon method and K-COS method, and we also realize both of them. We also proposed a new method for hand features points extraction(fingertips and finger valleys) based on Curvature scale space. This method get the position of fingertips and finger valleys by using gauss smooth on hand contour constantly and analyzing the local extreme of a modified curvature scale space(CSS) map which is obtained by adding curvature threshold. We compare our method with two state-of-the-art appearance-based approaches. Experimental results show that, although less quickly than two approaches, our method captures the hand articulations more precisely.(2)We realize the hand feature extraction system which is based on hand contour. This system take hand segmentation and extract the hand contour from the input KINECT depth image using open NI, and reposition the hand middle points by finding the biggest circle in hand contour. And obtain the fingertips and finger-valleys by using CSS method. Then we recover the fingertips of bending fingers which do not appear on the hand contour with a passable effect, using the extreme of depth derivative among points of special-selected region. Through a great number of experiments, we discuss and analyze the performance of our system. And we design a simple hand model to show our result.
Keywords/Search Tags:gesture recognition, hand segmentation, hand feature extraction, Curvature scale space
PDF Full Text Request
Related items