Real Time Fingertip Detection System Based On Convex Hull Analysis

Compared to traditional human-computer interaction which makes use of mice,keyboards, touch screen and so on, hand gestures recognition based human-computerinteraction offers a better choice that is non-contact, natural and friendly. Fingertipdetection (i.e., the position location) not only plays an important role in distinguishingdifferent hand gestures (especially gestures with high similarity), but also is the key toresponse to user’s fine manipulation accurately. However, accurate fingertip detectionremains a challenging task due to some negative factors, including complexbackground, varying illumination, different skin colors, various gestures, handmovement and so on.This paper focuses on detecting how many fingers (straight and not obscured)exit and locating their positions after the whole hand has been detected. The methodthat we propose consists of whole hand detection, adaptive segmentation of hand skinarea and fingertip detection based on convex hull analysis. If there is one hand in animage, then the hand can be detected and a rectangle enclosing the hand will beoutputted. Due to the fact that the outputted rectangle will contain part of thebackground (mainly the space between fingers and space around the hand), wepropose to segment the hand skin area through an on-line adaptive skin color methodwithin the rectangle in order to get rid of the unfavorable affect of non-skinbackground. After hand segmentation and subsequent hand contour extraction, convexanalysis is carried out on basis of both the whole and local characteristics of fingertipsbeing part of hand to find fingertip candidates and further remove false fingertips.The main contributions of this paper are as follows:(1) Hand segmentation usingan on-line adaptive skin color method is implemented within the rectangle containingthe hand, resulting in an improvement of hand segmentation.(2) We propose to detectfingertips through convex analysis based on both the whole and local properties of thehand.(3) We build a fingertip detection system with a head-mounted camera andverify the real-time property and accuracy of the proposed method.