基于肤色检测与卷积神经网络的手势识别

Hand gesture recognition based on feature fusion and convolutional neural network

针对光照变化、背景噪声等复杂环境对手势识别的影响,提出了一种基于YCb Cr空间肤色分割去除背景结合卷积神经网络进行手势识别方法。首先根据人体肤色在YCb Cr颜色空间中的聚类效果,采用基于椭圆模型的肤色检测方法进行手势分割;然后对分割后的手势图像提取骨架与边缘相融合的手势特征图;再通过深层次的Alex Net卷积神经网络结构,对经过融合的手势特征图进行识别。实验结果表明,针对复杂的背景环境,该算法具有较强的鲁棒性,在不同数据集下对手势的平均识别率提升了4%,可以达到99.93%。

To reduce the influence of the complex environment such as illumination changes and background noise to hand gesture recognition,an YCb Cr based space color segmentation model is used to remove background,and Alex Net convolution network is employed to achieve hand gesture classification in this paper. Based on the human skin color clustering in the YCb Cr color space,the method of skin detection based on elliptical model is used to extract gesture image. Then it extracts the gesture feature map fusing the skeleton and edge features from the segmented gesture images. Based on results of the features fusion of gestures,the Alex Net convolution network with deep level structure is employed to recognize the gesture image. The experimental results show that the proposed algorithm has strong robustness against the complex background environment,and the average recognition rate is improved by 4%,reaching to 99. 93%.