基于头姿识别的机器人轮椅智能交互系统

发布时间：2018-10-26 10:44

【摘要】：近年来,随着生活水平提高和医疗卫生技术的不断发展,残障和老年人士的生活质量受到了更多的关注,智能移动机器人的出现方便了他们的出行,成为了当前辅助器械医疗领域一个新研究热点。其中对于肢体残障人士,基于头部姿态的人机交互方式作为一个新的选择,受到了许多消费者和国内外科研人员的青睐,具有较高的市场价值与研究价值。本文在头部姿态识别研究的基础上,将头姿识别应用于移动服务机器人交互系统,实现了利用头部转动对移动服务机器人的运动控制。主要研究成果如下:(1)头部姿态识别是整个人机交互系统的基础,本文依据现有头部姿态研究方法,利用机器视觉算法对头部深度图像进行识别。本文使用Kinect深度传感器采集头部深度图像,由于原始深度图像存在噪声孔洞等缺陷,在进行识别时会存在诸多干扰,通过中值滤波算法和形态学滤波对原始深度图像进行预处理,方便后续的头姿识别。(2)根据当前国内外研究现状,利用随机森林和ICP算法对头部姿态进行计算,在分别实现两种算法后,分析两种算法各自在在精度与实时性方面的局限性。本文提出随机森林结合ICP算法的思想,利用随机森林算法作为ICP算法的粗配准,减少ICP算法在精配准阶段的迭代次数。两种算法互补对方的缺陷,使得头姿计算具备了实时性和精确性。(3)头部姿态与移动机器人运动状态之间的映射关系是人机交互系统的桥梁,本文提出头部朝向向量模拟操纵摇杆的控制方式,该方式符合人的自然交互方式,并通过实验验证具备良好的控制效果。根据以上分析开发基于头部姿态识别的智能移动服务机器人交互系统,以Kinect为图像采集装置,PC为计算控制核心,移动服务机器人为应用平台,实现了整个人机交互系统。
[Abstract]:In recent years, with the improvement of living standards and the continuous development of medical and health technology, more attention has been paid to the quality of life of the disabled and the elderly, and the emergence of intelligent mobile robots has facilitated their travel. It has become a new research hotspot in the medical field of auxiliary devices. As a new choice, the human-computer interaction based on head posture is favored by many consumers and researchers at home and abroad, and has high market value and research value. Based on the research of head attitude recognition, this paper applies the head attitude recognition to the mobile service robot interactive system, and realizes the motion control of the mobile service robot by using the head rotation. The main research results are as follows: (1) head attitude recognition is the basis of the whole human-computer interaction system. According to the existing head attitude research methods, this paper uses machine vision algorithm to recognize the head depth image. In this paper, the Kinect depth sensor is used to collect the head depth image. Due to the defects of the original depth image, such as noise holes, there will be a lot of interference in the recognition. The median filter algorithm and morphological filter are used to preprocess the original depth image to facilitate the subsequent head-pose recognition. (2) according to the current research situation at home and abroad, the head attitude is calculated by using stochastic forest and ICP algorithm. After the two algorithms are implemented, the limitations of the two algorithms in accuracy and real time are analyzed. In this paper, the idea of combining stochastic forest with ICP algorithm is presented. The random forest algorithm is used as the rough registration of ICP algorithm, and the iterations of ICP algorithm in the stage of fine registration are reduced. The two algorithms complement each other's defects, which makes the head-pose calculation real-time and accurate. (3) the mapping relationship between the head attitude and the moving state of the mobile robot is the bridge of the human-computer interaction system. In this paper, a control method of steering rocker with head orientation vector simulation is proposed, which is in line with the natural interaction mode of human beings, and is proved to have a good control effect by experiments. Based on the above analysis, an intelligent mobile service robot interaction system based on head attitude recognition is developed. The whole human-computer interaction system is realized with Kinect as the image acquisition device, PC as the computing control core and mobile service robot as the application platform.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【参考文献】