Static Hand Gesture Recognition Based On HOG Characters And Support Vector Machines

Gesture is another kind of communication which is daily used in life in addition to language, natural, understandable, and it is also an important media for transferring information. As an important subject of research in the field of human-computer interaction, the operating requirements of natural gesture recognition also represents the future direction of human-computer interaction: more natural, easier, low cost and user-friendly.In this paper, a thorough study on sensing operation of360-degree interactive experience District of Guangzhou Museum of cases,2008Shanghai World Expo is made, a new solution of gesture detection, feature extraction, classification is proposed on the basis of research of existing laboratory, aimed at improving the stability of the gesture recognition accuracy and recognition system.The main content and innovation of the paper include the following aspects:(1)Hand gesture detection and segmentation. A further improvement is made on the basis of laboratory’s existing research. The hand in the frame is segmented by means of combination of the prospect color segmentation and motion detection method, combines the advantages of both methods, compared to the previous method of simply using color segmentation, has made significant improvements in the segmentation results; exclude environmental aspects of skin color interference.(2)Hand gesture feature extraction. The HOG feature which was successfully used in pedestrian detection system was applied in the static hand gesture feature extraction. These feature descriptors used for target detection, count the times of direction of the gradient image appeared in local; this method is similar to the edge direction histogram, scale invariant feature transform and similar shape context method. Compared with other descriptors, HOG features maintain the geometry invariant, and has certain robustness for the gesture’s small amplitude rotation.(3)Hand-gesture classification recognition. This article uses a linear SVM classifier for classification. SVM is based on VC dimension theory and structural risk minimization principle on the basis of statistical learning theory, based on a limited sample of information-seeking among the most complex model (i.e., the accuracy of a particular learning training samples, Accuracy) and learning ability good compromise, with specific advantages in solving small sample size, nonlinear and high dimensional pattern recognition. In this paper, four kinds of custom gestures were classified, and achieved a higher recognition rate, and the system is more stable.(4)Realized the digital gesture recognition algorithm, and the algorithm is applied to gesture control PPT file playback system to achieve functions of opening, forward and backward flipping, turning off. The experiments show that, with more significant improvements in the recognition rate, the processing speed and stability gesture recognition algorithm based on static HOG.The innovation of this paper are mainly reflected in gesture detection, using a hybrid approach of color segmentation and motion detection, dividing the hand shape images with color and movement, effectively ruled out the skin color of the interference, to do the work for subsequent classification good bedding.