Research On Hand Gesture Recognition Based On Shape Features

As the main channel to share information and exchange intention between people, gestures play an extremely important role in human-computer interaction. With gestures, ordinary users can realize natural and friendly interaction with machines, allowing the machine to "understand" human language, and make appreciate actions. Combining with gesture shape features, this article mainly studies vision based gesture recognition algorithms from the perspective of natural and harmonious human-computer interaction. Aiming at the influence caused by gesture deformations on recognition accuracy, this paper realizes static and dynamic gesture recognition on basis of focusing on simple and effective gesture feature extraction research. Specific contents are as follows:A method of gesture recognition based on inner-distance shape context and bag of words (Inner-Distance Shape Context-Bag of Words, IDSC-BOW) is proposed in this paper, mainly aimed at the influence on the accuracy of recognition in representing process of gestures caused by joints or part structures deformations. Firstly, elliptical skin model is used to segment, get the binary gesture area and extract contours. Then, sample points on the contour are obtained by uniform sampling. The visual dictionary is generated through K-means clustering with inner-distance shape context features of hands. The generated visual dictionary is used to map the inner-distance shape context features of gestures into a collection of visual words. The BOW vectors are obtained by process of frequency statistic and normalization on the visual words. Finally, the support vector machine (SVM) classifier is used for classification. The experimental results show that the method has higher recognition rate on ten kinds of hand gestures as0-9. It keeps good robustness on joints of hands and part structures deformations.Aiming at the influence of joints or part structures deformations on the accuracy of gesture recognition, and large amount of calculation with shape matching directly, a method based on inner-distance contour point distribution features (IDCPDF) and histogram matching is proposed in this paper. Firstly, elliptical skin model is used to segment and extract contour. Then IDCPDF of gestures is generated. Finally, histogram matching is used to measure the similarity of IDCPDF and classify. Experimental results show that the method describes distributions of gesture contour points under polar coordinates. It not only reflects significant information of gesture shapes, but also reduces calculations in gesture features extraction and matching on the promise of ensuring gesture recognition accuracy, and achieves better real-time performance. Meanwhile, this method keeps good robustness on joints and part structures deformations of hands.