融合SURF与sEMG特征的手语识别研究  

Study on Sign Recognition Method Based on SURF and sEMG Feature Fusion

作　　者：林亚飞 曾晓勤[1] LIN Yafei;ZENG Xiaoqin(Institute of Intelligence Science and Technology, College of Computer and Information, Hohai University, Nanjing 211100)

机构地区：[1]河海大学计算机与信息学院,南京211100

出　　处：《微型电脑应用》2019年第4期55-57,共3页Microcomputer Applications

摘　　要：通过Kinect设备获取手语者手部的深度图像信息和彩色图像信息,采用阈值分割法有效滤除手语者所处环境中复杂的前景及背景信息;同时通过MYO臂环获取手语者的表面肌电信号信息以此来捕捉手语者微小的指尖动作信息,这样能够有效补充摄像头所拍摄不到的被遮挡的信息。把由Kinect获得的手语图像信息通过形态学处理方法提取其SURF特征后对其进行聚类生成合适大小的词袋模型(BOF-SURF),并用视觉词频向量表示手势语的局部特征,将其与MYO臂环获取的表面肌电信号(sEMG)特征进行融合,然后由SVM分类器通过五倍交叉验证的方法对手语库中手语进行学习和识别。实验结果表明,该方法既高效又具有很高的识别率,对30个中国手指语的最好识别正确率均可达97%左右。The Kinect device was used to obtain the deep image information and color image information of the signer’s hand, and the complex foreground and background information in the environment of the signer is effectively filtered through the threshold segmentation method. At the same time, the MYO arm ring is used to obtain the signal information of the signer to capture the tiny fingertip movement information of the sign language, it can effectively supplement the blocked information that the camera cannot capture. The sign language image information obtained by Kinect device through morphology processing methods is used to extract the SURF features, and then through clustering method to generate right size word bag model (BOF-SURF). The visual word frequency vector is used to represent the local characteristics of sign language, and then combining the local characteristics and EMG features to obtain MYO arm ring. The SVM classifier uses five-fold cross validation to learn and identify the sign language in the gesture library. The experimental results show that this method is effective and has high recognition rate, and the best accuracy rate of 30 Chinese gestures is 97%.

关 键 词：BOF-SURF KINECT SEMG SVM 手语识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]