基于特征提取的掌心实时定位与手势识别算法的研究
发布时间：2018-11-16 10:31
【摘要】：随着计算机视觉技术的深入发展,通过鼠标、键盘作为输入接口的传统交互模式已然无法满足人们的需求,用户直接与计算机交流的自然人机交互才是理想的人机交互模式。手势识别作为以人为中心的更自然的人机交互方式,能够满足用户与虚拟环境之间的直接交互,在智能家居、体感游戏、哑语识别等多个领域都有广泛应用。因此,实时手势识别技术具有重要的学术价值和应用前景。现有的手势识别研究仍然存在诸多不足:训练固定手势模板的识别方法难以满足实时性交互需求;单一基于手部肤色特征提取的手势识别方法在手势分割时容易出现分割不完整,导致识别率不高;基于可穿戴设备的手势识别方法要求用户必须佩戴数据手套等设备,此类设备价格昂贵且不利于推广。本论文主要针对手势识别算法中的手势区域肤色分割、手势特征提取、数字手势识别三大部分进行了详细研究,主要工作如下:首先在手势区域分割阶段,研究了普通摄像头平台下采集图像未出现人脸和出现人脸两种情况下的分割方法。当摄像头采集图像未出现人脸时通过阈值化肤色检测器直接提取手势区域;当摄像头采集图像出现人脸时,根据人脸与手部肤色一致性原则,提出了融合人脸肤色检测的手势区域分割新方法。该方法通过人脸检测获取人脸区域肤色像素值范围,再与传统阈值法相结合,通过双阈值法准确的分割背景和手势区域,可以在一定程度上优化分割效果。其次在手势特征提取阶段,采用运动目标图像检测方法,在两帧帧间差分法的基础上,采用三帧帧间差法结合肤色分割来实现运动手势区域检测。根据本文算法提取了手势轮廓、手势凸包与凸缺陷、指尖与指间凹槽等重要手势特征,并在此基础上提出了基于指间凹槽最小外接圆的实时掌心检测定位方法,准确有效的实现了手势掌心的定位。该方法适应手势区域进行平移、旋转、翻转等不同场景下的识别特征提取要求,具有较好的鲁棒性。最后在数字手势识别阶段,根据提取到的指尖与指间凹槽和掌心等特征,构造出识别决策树模型,实现了对常用的数字手势快速准确识别。搭建的基于VS 2013集成开发平台的实时手势识别系统验证了上述方法的有效性。
[Abstract]:With the development of computer vision technology, the traditional interaction mode of mouse and keyboard as input interface can no longer meet the needs of people. The natural human-computer interaction between users and computers is the ideal human-computer interaction mode. Gesture recognition, as a more natural human-computer interaction mode, can satisfy the direct interaction between users and virtual environment. It has been widely used in many fields such as smart home, body sense game, mute recognition and so on. Therefore, real-time gesture recognition technology has important academic value and application prospect. The existing research on gesture recognition still has many shortcomings: the recognition method of training fixed gesture template is difficult to meet the real-time interaction requirements; Hand gesture recognition method based on hand skin color feature extraction is prone to incomplete hand gesture segmentation, which leads to low recognition rate. Gesture recognition based on wearable devices requires users to wear data gloves and other devices, which are expensive and unsuitable for promotion. In this paper, the skin color segmentation of gesture region, gesture feature extraction and digital gesture recognition are studied in detail. The main work is as follows: firstly, in the phase of gesture region segmentation, In this paper, the segmentation method of image without face and human face is studied on the common camera platform. When the face is not seen in the image captured by the camera, the gesture area is directly extracted by the threshold skin color detector. According to the principle of consistency between the face and the skin color of the hand, a new method of hand gesture region segmentation based on facial color detection is proposed. This method obtains the range of skin color pixels of face region by face detection and combines with the traditional threshold method. By using double threshold method to segment the background and gesture regions accurately the segmentation effect can be optimized to a certain extent. Secondly, in the phase of gesture feature extraction, the moving target image detection method is adopted, and based on the difference method between two frames, the three-frame inter-frame difference method combined with skin color segmentation is used to realize the motion gesture region detection. According to this algorithm, some important gesture features, such as gesture contour, gesture convex hull and convex defect, finger tip and finger grooves, are extracted. Based on this, a real-time palm detection and localization method based on the minimum circumscribed circle of interdigital groove is proposed. Accurate and effective hand gesture palm positioning. This method adapts to the requirements of feature extraction in different scenes, such as translation, rotation and flipping, and has good robustness. Finally, in the phase of digital gesture recognition, a decision tree model is constructed according to the extracted features such as grooves and palms between fingertips and fingers, and the recognition of common digital gestures is realized quickly and accurately. A real-time gesture recognition system based on VS 2013 integrated development platform is built to verify the effectiveness of the above method.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
[Abstract]:With the development of computer vision technology, the traditional interaction mode of mouse and keyboard as input interface can no longer meet the needs of people. The natural human-computer interaction between users and computers is the ideal human-computer interaction mode. Gesture recognition, as a more natural human-computer interaction mode, can satisfy the direct interaction between users and virtual environment. It has been widely used in many fields such as smart home, body sense game, mute recognition and so on. Therefore, real-time gesture recognition technology has important academic value and application prospect. The existing research on gesture recognition still has many shortcomings: the recognition method of training fixed gesture template is difficult to meet the real-time interaction requirements; Hand gesture recognition method based on hand skin color feature extraction is prone to incomplete hand gesture segmentation, which leads to low recognition rate. Gesture recognition based on wearable devices requires users to wear data gloves and other devices, which are expensive and unsuitable for promotion. In this paper, the skin color segmentation of gesture region, gesture feature extraction and digital gesture recognition are studied in detail. The main work is as follows: firstly, in the phase of gesture region segmentation, In this paper, the segmentation method of image without face and human face is studied on the common camera platform. When the face is not seen in the image captured by the camera, the gesture area is directly extracted by the threshold skin color detector. According to the principle of consistency between the face and the skin color of the hand, a new method of hand gesture region segmentation based on facial color detection is proposed. This method obtains the range of skin color pixels of face region by face detection and combines with the traditional threshold method. By using double threshold method to segment the background and gesture regions accurately the segmentation effect can be optimized to a certain extent. Secondly, in the phase of gesture feature extraction, the moving target image detection method is adopted, and based on the difference method between two frames, the three-frame inter-frame difference method combined with skin color segmentation is used to realize the motion gesture region detection. According to this algorithm, some important gesture features, such as gesture contour, gesture convex hull and convex defect, finger tip and finger grooves, are extracted. Based on this, a real-time palm detection and localization method based on the minimum circumscribed circle of interdigital groove is proposed. Accurate and effective hand gesture palm positioning. This method adapts to the requirements of feature extraction in different scenes, such as translation, rotation and flipping, and has good robustness. Finally, in the phase of digital gesture recognition, a decision tree model is constructed according to the extracted features such as grooves and palms between fingertips and fingers, and the recognition of common digital gestures is realized quickly and accurately. A real-time gesture recognition system based on VS 2013 integrated development platform is built to verify the effectiveness of the above method.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
【相似文献】
相关期刊论文 前10条
1 武霞;张崎;许艳旭;;手势识别研究发展现状综述[J];电子科技;2013年06期
2 ;新型手势识别技术可隔着口袋操作手机[J];电脑编程技巧与维护;2014年07期
3 任海兵,祝远新,徐光，
本文编号：2335268
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2335268.html
