期刊名称：IAENG International Journal of Computer Science
印刷版ISSN：1819-656X
电子版ISSN：1819-9224
出版年度：2018
卷号：45
期号：4
页码：569-583
出版社：IAENG - International Association of Engineers
摘要：In the hand gesture recognition process, manuallydesigned features are difficult to achieve good results under thecondition of changeable gestures and complex backgrounds. Inthis paper, we propose a hand gesture recognition method basedon Gaussian skin color model and deep convolutional neuralnetwork (DCNN). For gesture images in different backgrounds,we first use the Gaussian skin color model to segment the gesturearea, then we use the DCNN to establish gesture classificationmodel. Finally, we use the back propagation algorithm based onpartial differential equation to train the neural network on thepure gesture data samples to converge to the global optimum,and obtain the classification results. The model combines theprocess of feature extraction and classification, simulates thebiological visual transmission and cognition, and effectivelyavoids the subjectivity and limitations of artificial features. Andmodel reduces the size and the complexity of network by usingweights sharing and pooling technology. Experimental resultsshow that the method is efficient for gesture representation andclassification. The average classification accuracies under twodatasets (indoor and outdoor environments) are both more than99%. Compared with the traditional methods, the proposedmethod has higher classification accuracy and speed.
