Research Of Gesture Recognition Algorithm Based On Fingertips Location

With the rapid development of the computer, the emerged new human-computer interaction technologies are more natural and efficient. Gesture is one of the basic ways of communication and accords with human habits of daily life. As the active way of human-computer interaction, the research of gesture recognition is important. But now the actual application occasions are not too much, most of the gesture recognition algorithms are still in research stage. In view of the above mentioned, static gesture recognition and dynamic trajectory recognition are studied in this paper.This paper presents a new real-time gesture recognition algorithm. Static gesture recognition system can identify 30 Chinese sign language letters, the average recognition rate is 93.7%. The dynamic trajectory recognition can recognize five kinds of trajectory of the hundred, thousand, arc,% and CO2 in Chinese sign language, the average recognition rate is 85.1%.In the part of skin color segmentation, this paper puts forward an adaptive skin color detection algorithm which combines with background difference. Firstly a new HSV-CbCr mixed color space is set up to detect the skin region in it. According to background difference, the area of hand is left. Then the accurate skin information is got to adjust the color detection threshold parameters. This is an adaptive algorithm of skin color segmentation. The algorithm can adapt to the skin color difference of different users and avoid the fluence of skin color area in background. In the mixed color space, the skin color’s clustering and distinguishing are took into account and the influence of illumination changes is reduced.This paper presents a fingertip detection algorithm that combines the convex hull and curvature analysis of gesture’s contour. The fingertips are located in the convex hull vertices. First of all, the algorithm narrow down the candidate points of fingertips through the calculation of convex hull. Then the candidate points of fingertips can be reduced further based on the convex defects depth. Finally, the program computes the candidate points of contour curvature to remove stubborn interference and get the exact location of the fingertips. The single fingertip localization is realized by convex hull analysis. The dynamic trajectory of gesture is given by the connecting of single fingertip localizations in chronological order.In the feature extraction of static gestures. the gesture’s structure feature and statistical feature are selected to form a feature vector. These features not only have the characteristics of generality and independence, but also have low computation complexity and high efficiency. The feature of dynamic trajectory include the angle changing in quantitative encoded 12 direction and structure characteristics of the trajectory.Based on the feature extraction, the multi valued classifier of Support Vector Machine (SVM) is built by the way of one-by-one. After training by sample, the classifiers can recognize the static gesture and dynamic trajectories.