结合手势二进制编码和类-Hausdorff距离的手势识别

发布时间：2018-05-08 23:02

本文选题：动态手势识别 + 手势主方向　；参考：《电子学报》2017年09期

【摘要】：针对目前动态手势识别方法受手势旋转、平移、缩放的影响,并解决手势识别的实时性问题,提出一种基于手势二进制编码和类-Hausdorff距离模板匹配的手势识别方法.首先,把分割好的手势图像进行标准化处理,求出标准化图像中的手势主方向,建立二维手势直角坐标系,提取空间手势特征;其次,根据前五帧手势图像中手势像素点个数的变化量识别出动态手势类型;然后,用手势二进制描述子从动态手势类型中再筛选出可能的候选手势集合;最后,用类-Hausdorff距离模板匹配方法从候选手势集合中识别出最终手势.主要创新点在于:提出的动态手势类型识别和手势二进制描述子匹配的方法,大大缩短了动态手势识别的时间;提出的结合手势主方向的类-Hausdorff距离方法,不仅对旋转、平移和缩放手势具有不变性,而且对区分度较小的手势也具有较高的识别准确率.实验结果表明,在光照相对稳定的条件下,该方法能够实时准确的实现动态手势识别,总体识别率达到95%以上,对发生缩放的手势识别率能达到92%以上,对发生旋转的手势识别率能达到87%以上.本文算法已经在一个基于手势的人机交互界面中得到应用.
[Abstract]:Aiming at the influence of rotation, translation and scaling of gesture, and solving the real-time problem of gesture recognition, a gesture recognition method based on gesture binary coding and matching of -Hausdorff distance template is proposed. First of all, the segmented gesture image is standardized, the main direction of the gesture in the standardized image is obtained, the 2D gesture coordinate system is established, and the spatial gesture features are extracted. The dynamic gesture types are identified according to the number of gesture pixels in the first five frames of gesture images. Then, the possible candidate gesture sets are selected from the dynamic gesture types using gesture binary descriptors. The final gesture is identified from the candidate gesture set by using the -Hausdorff distance template matching method. The main innovation lies in: the proposed method of dynamic gesture type recognition and gesture binary description sub-matching greatly shortens the time of dynamic gesture recognition, and the proposed method combines the main direction of gesture with the class -Hausdorff distance method, which is not only for rotation, but also for dynamic gesture recognition. The translation and scaling gestures are invariant, and the recognition accuracy is higher for the less discriminant gestures. The experimental results show that under the condition of relatively stable illumination, the method can realize dynamic gesture recognition in real time and accurately, the overall recognition rate is more than 95%, and the recognition rate for zoom gesture can reach more than 92%. The recognition rate of the rotation gesture can reach more than 87%. This algorithm has been applied in a gesture-based human-computer interface.
【作者单位】：济南大学信息科学与工程学院;山东省网络环境智能计算技术重点实验室;山东省分布式计算机软件新技术重点实验室;
【基金】：国家自然科学基金项目(No.61472163,No.61603151) 国家重点研发计划项目(No.2016YFB1001403) 山东省重点研发计划项目(No.2015GGX101025) 济南大学博士科研基金(No.XBS1534)
【分类号】：TP391.41

【相似文献】