摘要
针对手势识别过程中分割出的手势不精确、利用单特征识别时识别率低等问题,提出一种使用深度信息进行多特征提取的手势识别算法。利用Kinect得到深度信息并完成人手定位,将手部区域细分成手掌区域、指尖区域和手臂区域;提出3个不同的特征描述子,即指尖点到手掌中心点的距离、指尖点到手掌平面的距离以及手掌区域特征;应用一个多分类的支持向量机(SVM)分类器对手势进行分类,在所建手势数据库中完成算法验证。实验结果表明,该算法能够精确分割手部区域,手势识别率得到很大提高。
A kind of recognition algorithm which used depth information to process feature extraction was proposed, for a series of problems in the process of gesture recognition, for example, the gesture divided is not accurate and the recognition rate is low when using single feature recognition. The positioning of the hand was acquired according to the depth information obtained using Kinect, and the hand regional area was divided into palm area, fingers area and arm area. Three different feature descriptors were put forward including the distance from fingertips to the center of the palm, the distance from fingers to the palm plane and the geometrical characteristics of the palm area. The algorithm validation was completed in the gestures database that built by applying a multi-class support vector machine classifier to classify gestures. The experimental results show that the algorithm can finish the gesture recognition quickly, and the recognition rate is improved significantly.
出处
《计算机工程与设计》
北大核心
2017年第4期953-958,共6页
Computer Engineering and Design
基金
山西省国际科技合作计划基金项目(2014081012)
山西省科技攻关基金项目(2015031003-3)
关键词
深度图像
视觉特征
多特征提取
SVM分类
手势识别
depth image
visuaI features
multi-feature extraction
support vector machine classifier
gesture recognition
