基于深度学习多特征融合的手势分割识别算法

Gesture Segmentation and Recognition Algorithm Based on Deep Learning Multi-feature Fusion

针对机器视觉技术因受环境中类肤色物体、背景变化及手势不同角度的影响而导致手势分割精度和识别准确度较低的问题,提出融合肤色结合深度学习网络的手势分割识别算法;该算法根据手势肤色在颜色空间中的聚类特征,融合颜色空间YC_(b)C_(r)、HSV对肤色聚类较好的色调、蓝色色度、红色色度3个分量,结合最大类间方差的自适应提取方法,对手势部分进行分割处理,并在更深层次的深度学习网络中对手势特征图像进行训练。结果表明,该算法能较准确地分割和识别手势,平均识别率为98.8%,具有较强的鲁棒性。

In view that machine vision technology was affected by skin color like objects in the environment,background changes,and different angles of gestures,resulting in low accuracy of gesture segmentation and recognition,a gesture segmentation and recognition algorithm based on skin color and deep learning network was proposed.According to clustering characteristics of gesture skin color in color space,the algorithm combined 3 components of hue,blue chroma,and red chroma which color spaces of YC_(b)C_(r)and HSV clustered better for skin color,and combined the adaptive extraction method of maximum class variance to segment the gestures partly.Gesture feature images were trained in deep learning network at a deeper level.The results show that the algorithm can segment and recognize gestures accurately with the average recognition rate of 98.8%,and has strong robustness.