A fast static gesture recognition method

摘要

Gesture recognition has become a hot spot in human interaction technology. But the difference of skin colour, the complexity of the background and the rotation of the gesture make it difficult to complete gesture segmentation and recognition. To overcome the impact of the difference of skin colour, in this paper gesture images are split out based on skin colour in the hue saturation value (HSV) colour space combined with the mean-shift algorithm. Then the Hu invariable moments of the gesture images are calculated as the feature vector to overcome the impact of the rotation of gesture. In the experiment 660 images were collected from 10 experimenters. Three hundred and thirty images of those were made up by left-hand gestures and the rest were made up by right-hand gestures. The experimental results show the accuracy of this algorithm is from 90% to 100% with single-hand static gestures.