鱼眼图像中手部定位及手势识别

摘要

In view of the demand for hand gesture interaction in Augmented Reality (AR), this paper proposes a method of hand location and hand gesture recognition in fish-eye image.First, skin color region is detected in YCbCr color space, then HSV space color histogram feature and shape feature is used to detect mark in the skin region, and to divide hand area.Second, this method uses the centroid of mark as hand positioning center, according to imaging principle of fish-eye lens, to calculate the 3D position of the point relative to fish-eye lens.Last, this paper extracts hand gesture shape features, to complete the procedure of recognizing hand gestures.Through the experiment of the 85 fish-eye images with different background and light, the correct rate for recognizing gestures is 96％, recognizing single fish-eye image takes about 0.03 s.The experimental results show that the proposed method can realize hand location and hand gesture recognition quickly and effectively in complex background.%针对增强现实领域手势交互的需求,提出鱼眼图像中手部定位及手势识别的方法.首先在YCbCr颜色空间中检测肤色区域,再利用HSV空间颜色直方图特征和形状特征在肤色区域中检测标识物,分割手部区域.然后将标识物质心作为手部定位中心,根据鱼眼镜头的成像原理,计算该点相对于鱼眼镜头的三维位置.最后提取手势形状特征,实现手势识别.通过对85幅不同背景及光照条件下的鱼眼图像进行实验,此方法识别手势的正确率为96％,识别单幅鱼眼图像约耗时0.03 s.实验结果表明,该方法可以快速准确地实现在复杂背景下的手部定位与手势识别.