【作者】 李佳慧；

【导师】 钟慧湘；

【作者基本信息】 吉林大学 ， 计算机应用技术， 2016， 硕士

【摘要】 计算机视觉的快速发展,使基于计算机视觉的人机交互技术也在随之进步。利用传统上的鼠标、键盘等物理输入媒体的人机交互方式已经不能够满足人们的需求。所以更快更便捷的人机交互方式成为现今研究的主流课题,通过计算机视觉识别相关人体动作,并利用程序计算使计算机给出相应反馈动作这是人们想要得到的人机交互最终目的。利用手部动作来实现人机交互是目前为止最为方便可行的方法之一,基于计算机视觉的手势交互系统因此被提出,这项技术在体感游戏、智能家居等方面均有相关应用。指尖的检测是手指交互系统的关键所在,本文利用指尖位置检测实现对鼠标移动的控制。指尖检测的研究主要有二维和三维空间两个方向,当然三维空间因为采集数据种类比较丰富,判定条件增加,因此识别率也会较高,但是由于手部运动过程数据量较大,所以这种方法的实时性也会随之降低,实用性与便捷性较差,所以本文研究重点放在二维单目视觉下的指尖检测上。指尖检测可划分为动态检测与静态检测两种,动态检测时要进行背景分割将作为手部的前景提取出来,因为均值背景消减法的实时性较好,固定镜头下的分割效果较好,所以在分离背景本文选用此方法。然后利用肤色判断作为分割条件,本文通过对比实验以及检测肤色聚集度两种形式确定肤色模型,选取YCb Cr肤色模型用作肤色分割,但是由于判断肤色的像素阈值不能够自适应,所以仍然存在分割效果不好的情况,因此本文提出改进的圆形梯度算法,能够在分割不好的手形上设置种子点,利用相关条件来填补肤色检测后的漏洞位置,更好的完成手部分割。经过形态学相关处理后,本文使用canny算子可以提取到圆润手部轮廓,为接下来的指尖准确检测做下良好的铺垫。接下来本文使用改进后能够自适应提取K值的指尖检测算法将指尖位置准确检测出来。并利用实验检测与对比,验证本文算法的有效性。最后利用检测出的指尖位置与鼠标信息相结合对鼠标进行操控,实现对算法的简单应用。更多还原

【Abstract】 With the rapid development of computer vision, the human-computer interaction technology based on computer vision is accordingly progress. By using the traditional mouse, keyboard and other physical input media, human computer interaction mode has not been able to meet the needs of people. So the faster and more convenient way of human-computer interaction become mainstream topics of current research, using computer vision to extract human action, and using program to calculate then the computer will gives the corresponding feedback action, which is the ultimate goal of human-computer interaction.Among them, communicating with computer by hand is the most convenient and feasible method of that human-computer interaction. So human-computer interaction based on computer vision has been proposed. It is used in somatosensory games, smart home and some other. The fingertip detection is the key of finger interaction system, this paper uses the fingertip position to control the movement of the mouse. Fingertip detection research has two directions: two-dimensional and three-dimensional space. Of course, in three-dimensional space there are many data types detecting fingertip, so the recognition rate will be higher in this space. But at the same time, it’s rich data may take out consume more time, and practicability and convenience are poor, practicality and convenience are poor, so this paper focuses on the detection of fingertip under two-dimensional monocular vision. Fingertip detection can be divided into two kinds of dynamic and static, dynamic detection should be carried out as a background segmentation in which hand foreground is extracted. Because the mean of background subtraction method is more real-time, and a good result of segmentation under fixed lens camera, so we choose it. And then we use skin color as a judgment of segmentation, this paper through two ways to determine skin color model, the comparison experiment and the skin color clustering. Then we choose YCb Cr skin color model for the image with threshold skin pixels are extracted, but because the threshold cannot be adaptive, so there are still segmentation effect is not good. Therefore, we use the circular gradient algorithm, which can set seed in the shape of hand segmentation and fill the holes of skin color detection, and get a better completed hand segmentation. After morphological processing, we use the Canny operator to extract the rounded hand contour, which prepare for the next fingertip accurate detection. Then, we use the improved algorithm to extract the K value of the fingertip detection algorithm to detect the fingertip position accurately. We do many experiments to prove effectiveness of the proposed algorithm in the paper. Finally, using the detected fingertip position and the mouse information to control the mouse, and achieve a simple application of the algorithm.更多还原