基于视觉手势识别的机械手操控系统的研究

发布时间：2018-01-08 01:15

本文关键词：基于视觉手势识别的机械手操控系统的研究　出处：《天津工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着机器人技术的飞速发展,机器人已经广泛地应用于生产、生活等各个领域中,人与机器人之间的交互活动也就变得越来越频繁。但是目前机器人的操控方式却很单调、繁琐,人与机器人之间的交互不够直观、自然而且效率比较低。因此,研发基于视觉的自然手势交互系统不仅具有一定的学术研究价值,而且具有相当大的实际应用前景。本文在分析了国内外手势识别和机器人控制系统的相关研究资料及成果的基础上,选取了 Leap公司出产的Leap Motion传感器获取人手信息,并针对单个传感器使用时因视觉干扰或者不可避免的遮挡而导致手势识别准确率降低的现象,研究并实现了一种基于多Leap Motion传感器的机械手手势操控系统。这一系统采用多个传感器在多方位对操作者的手势进行探测追踪并进行手势识别,实现了通过自然手势控制机械手动作的非接触式交互。该系统主要包括客户端(手势检测模块)、信息服务器端(数据处理模块)、机械手控制器及机械手机构,文中主要对手势采集模块以及数据处理模块进行了研究讨论。该文首先介绍了该基于手势识别的机械手工作原理,然后在对Leap Motion传感器进行了一定的研究分析后,确定了多个传感器位置布局,实现了一个基于多传感器的手势数据采集模块。该系统中数据处理模块的主要作用是对从多个传感器获取的手势数据进行数据分析处理,为了更加及时地响应人手运动状态的变化和获取较稳定、准确的人手位置数据,本文提出了一种改进的基于"当前"统计模型的自适应卡尔曼滤波算法进行滤波和轨迹追踪;对于通过手指指骨向量计算得到的关节角度,采用一种改进的基于均值的加权滑动均值滤波算法进行稳定性滤波平滑处理。当进行手势的姿态识别时,将手指关节角度等手势特征向量送入支持向量机(SVM)进行手势分类,并采用交叉编译和网格搜索的方法对其参数进行寻优,手势识别准确率可达98.5%。数据处理模块会实时地将位置增量、旋转角度增量或者姿态动作指令发送给机械手控制器,机械手控制器会通过内部运动学反解算法得到机械手各个关节轴运动数据从而驱动各个伺服电机转动。在最后的系统实验环节,操作者通过改变自身手部的空间位置和姿态达到了手势操控机械手运动的目的,并验证了该系统的可用性以及数据处理算法的有效性。
[Abstract]:With the rapid development of robot technology, the robot has been widely used in production, life and other fields, the interaction between human and robot becomes more and more frequent. However, the robot is very monotonous, tedious, interaction between human and robot is not intuitive, natural and relatively low efficiency. Therefore, research and development of natural gesture interactive system based on vision not only has a certain academic value, but also has considerable practical application prospect. Based on the analysis of relevant research data and achievements at home and abroad for gesture recognition and robot control system on the selected Leap Motion sensor produced by the Leap company staff get information, and for a single sensor when used for shielding the visual interference or inevitably caused the gesture recognition accuracy decreases, and the implementation of a study Mechanical hand gesture control system based on Motion sensor Leap. Multiple sensors on the operator's gesture in multi azimuth detection tracking and gesture recognition using this system, through the implementation of contactless natural interactive gesture control manipulator movement. The system includes client (gesture detection module, server (information) the data processing module), mechanical manipulator controller and mobile phone structure, this paper focuses on the gesture acquisition module and data processing module are discussed. This paper first introduces the mechanical hand gesture recognition based on the principle of work, and then in the Leap Motion sensor was analyzed after certain to determine the location of layout a sensor, to achieve a gesture based on multi sensor data acquisition module. The main function of data processing module in the system is obtained from multiple sensors The gesture data for data analysis and processing, in order to more timely response to changes in the hand motion state and obtain a stable, accurate hand position data, this paper proposes an improved based on the current statistical model and adaptive Calman filtering algorithm is used for trajectory tracking; joint angle obtained by finger phalanx vector calculation. Using an improved weighted moving average filtering algorithm based on mean stability smoothing. When the attitude recognition of gestures, finger joint angle gesture eigenvector into support vector machine (SVM) method is adopted for gesture classification, cross compiler and grid search to optimize the parameters of gesture recognition the accuracy of 98.5%. data processing module in real time position incremental, rotation angle increment or gestures instructions sent to the manipulator control The controller, the robot controller through the internal inverse kinematics algorithm of manipulator joints motion data obtained thereby each servo motor rotation. In the last part of the experiment system, the operator by changing the spatial position and posture of the hand to motion gesture control manipulator, and validate the availability of the system and data processing algorithms.

【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【相似文献】