【作者】 刘昌盛；

【导师】 厉树忠；

【作者基本信息】 西北师范大学 ， 电路与系统， 2008， 硕士

【摘要】 随着我国计算机技术的迅猛发展,人们对计算机使用方便程度的要求越来越高,手势是一种自然而直观的人际交流模式。已经成为一种重要的人机交互方式。基于视觉的手势识别是实现新一代人机交互所不可缺少的一项关键技术。然而,由于手势本身具有的多样性、多义性、以及时间和空间上的差异性等特点,加之人手是复杂变形体及视觉本身的不适定性,因此基于视觉的手势识别是一个极富挑战性的多学科交叉研究课题。手势分为动态手势和静态手势,动态手势定义为手运动的轨迹,而静态手势强调通过手型传递一定的意义。本文从手势的分割、手势的特征提取和识别三个方面对静态手势识别算法进行了研究。手势的分割是所有手势识别系统的第一步也是最为重要的一步,分割的效果直接影响到后续的识别结果。本文提出了一种新的基于LM_BP神经网络的手势分割方法。这种方法不需要进行色彩空间的转换,不需要考虑肤色在色彩空间的聚类效果,只要训练样本足够丰富,便可以通过神经网络对肤色在颜色空间的分布进行精确的描述。训练后的网络可以将人体的肤色信息从复杂背景中较好的分割出来。在手势图像的特征提取和识别部分,本文首先采用八邻域搜索法对二值化的手势图像进行边缘检测,得到连通的手势外轮廓,然后分析了具有旋转、平移和尺度变换不变性,并且与边界的起点位置无关的归一化傅立叶描述子,并把归一化傅立叶描述子及欧式距离应用于字母手势的识别中,通过计算输入手势的归一化傅立叶描述子与样本库中各类图像的特征向量的欧式距离,判定输入图像与样本图像间的匹配程度,把待识别的输入图像归为距离最小的那一类。实验结果表明,提出的方法对复杂背景中的字母手势识别率可达到88.46%,可实时识别。更多还原

【Abstract】 with the computer technique developed,hand gesture plays a natural and intuitive communication mode for all human dialogs. Hand gestures play a natural and intuitive communication mode for all human dialogs. The ability for computer to visually recognize hand gestures is essential for future human-computer interaction(HCI). However ,vision-based recognition of hand gestures is an extremely challenging interdisciplinary project due to following three reasons :(1) hand gestures are rich in diversities ,multi-meanings ,and space-time varieties ; (2) human hands are complex nonrigid objects ; (3) computer vision itself is an ill-posed problem.There are two kinds of gesture, dynamic gesture which has the definition as movement locus of hands and static gesture which expresses meaning through hand-shape. This paper tries to perform study on static hand gesture recognition including three parts, hand image segmentation, extracting hand features and recognition.Hand image segmentation separates the hand image from the background. It is the first important step in any hand gesture recognition system, and all subsequent stages heavily rely on the quality of segmentation. In this paper, a novel color segmentation approach for hand image segmentation is developed on the basis of Levenberg-Marquardt Back Propagation(LM_BP) neural network. In this method, there is no need to transform color space and no need to think of the distribution region of skin colors in a color space.With the abundant training samples, the LM_BP network is capable to characterize the distribution region of skin colors accurately in the color space and identify the skin regions efficiently in the color image with complex backgrounds.In the part of feature extraction and recognition, we get the connected gesture contour in the way of edge detection based on 8-connected boundary tracking,then the paper analyzes normalized Fourier Descriptors which do not change with translation,rotation and scale change, and which are also independent of the location of the beginning point of the image edge. The normalized Fourier Descriptors and Euclidean Distance are applied to the recognition of the alphabet gesture. We can judge the resemblance between input image with each image in database by computing the distance between their feature vectors, and we classify the image into the sample class that has the shortest distance.The experiment result shows that the alphabet hand posture recognition ratio with complex backgrounds is 88.46% and the real-time recognition is capable.更多还原