摘要
为实现基于Kinect的手语识别,提出了一种利用有限状态机及动态时间规整(DTW)的动态手语识别方法。首先,利用Kinect技术得到人体深度图像和骨骼特征信息;然后利用手部分割算法得到手部深度图像,再选取识别正确率高的梯度方向直方图(HOG)特征算子来提取手部特征;最后加入有限状态机和DTW算法实现动态手语识别。实验结果表明:该方法能够实现对常用手语单词、句子的识别,识别准确率可达95%。
In order to realize sign language recognition based on Kinect,a dynamic sign language recognition method using finite state machines and dynamic time warping(DTW)is proposed.Firstly,the Kinect technology is used to get the depth image of human body and skeleton feature information.Next,the hand depth image is obtained by using the hand segmentation algorithm.And then,the characteristics of the hand are extracted by selecting the histogram of oriented gradient(HOG)feature operator with high accuracy of recognition.Finally,the finite state machine and DTW algorithm are added to realize dynamic sign language recognition.The experimental results show that the method can achieve the recognition of normal sign language words and sentences,and the recognition accuracy can reach up to 95%.
作者
千承辉
邵晶雅
夏涛
刘怀宾
QIAN Chenghui;SHAO Jingya;XIA Tao;LIU Huaibin(College of Instrumentation&Electrical Engineering,Jilin University,Changchun 130000,Ch)
出处
《传感器与微系统》
CSCD
2019年第6期31-34,38,共5页
Transducer and Microsystem Technologies
基金
国家级大学生创新实践基金资助项目(2016A65281)
