融合深度信息和稀疏自编码的手势识别算法

The Gestures Recognition Algorithm Fusing Depth Information and Sparse AutoEncoder

针对目前深度手势识别算法无法有效去除手腕区域干扰且识别率较低的问题,提出了一种融合深度信息和稀疏自编码的手势识别算法。改进算法利用深度信息和肤色信息相结合的方法对样本图像进行粗分割,对分割后的图像采用手势端点检测和可变阈值算法进行精确分割,去除人体手腕区域的干扰,得到精确的手势分割区域。然后将分割后的样本图像输入稀疏自编码神经网络中进行无监督的训练与特征提取,并将提取的特征输入到softmax分类器中进行识别,得到手势的识别结果。仿真结果表明,改进算法能够有效去除手腕区域部分,且具有更高的识别率。

Aiming at the problems that the current depth gestures recognition algorithms cannot effectively remove the interference of wrist area and low recognition rate,the paper proposes a gestures recognition algorithm which fuses depth information and sparse autoencoder. The improved algorithm uses the combination of depth information and color information to coarsely split the sample image. Then it uses gesture endpoint detection and variable threshold algorithm to accurately split the image after coarsely segmentation. By this way,it can remove the interference of wrist area and get the precise gesture segmentation area. Then the segmented sample is input into the sparse autoencoder neural network for unsupervised training and feature extraction. The extracted features are input into the softmax classifier for identification,and the recognition result of the gesture is obtained. The simulation result shows that the improved algorithm can effectively remove the region of the wrist and has higher recognition rate.