仿生眼三维感知与运动控制研究

CASIA OpenIR > 毕业生 > 博士学位论文

	仿生眼三维感知与运动控制研究
	王庆滨
	2017-05-23
学位类型	工学博士
中文摘要	仿生眼是智能机器人系统中必不可少的组成部分。目前，国内外关于仿生眼的研究主要集中于仿生运动和视觉感知机理方面，但如何将目标三维信息与仿生眼的仿生运动控制进行有机结合，还有待进一步探讨。因此，本文针对仿生眼的三维感知和基于三维信息的仿生运动控制展开研究，着重研究仿生眼的三维感知、扫视和平滑追踪运动控制、机器人原位注视点跟踪和趋近注视控制。首先，设计了一套4自由度的仿生眼系统。该系统通过双目CCD摄像机和视频采集卡获取图像。两台相机分别被安装在左右运动模块上。运动模块通过两个步进电机实现左右旋转和上下俯仰。步进电机的角度通过旋转编码器采集得到。基于该仿生眼系统，提出了一种单双目相结合的三维感知方法。用手眼标定方法得到相机坐标系与运动模块坐标系的变换关系。利用初始位置双目的外参数、双目的手眼参数和步进电机的角度可以得到任意时刻双目的外参数。根据双目外参数、目标在左右图像中的坐标和双目相机内参数，可以计算得到目标的三维坐标。根据不同时刻单目的外参数、目标的图像坐标和相机内参数，利用单目多视图融合的方法得到目标的三维坐标。通过单目运动立体视觉和双目立体视觉的结合可以扩大仿生眼的感知视野。其次，针对仿生眼系统，提出了一种基于视觉和限位开关结合的廉价高精度初始定位方法。当系统启动时，第一步利用限位开关进行粗定位，第二步基于最小二乘椭圆拟合进行视觉精定位。这种方法价格低廉且定位精度高，平台每次启动不需要重复标定双目初始位置的外参数。提出了一种基于主动目标注视的立体视觉测量方法。通过离线标定得到双目的基线和相机的运动角度偏差。在测量目标的三维坐标时，一个相机固定不动，作为监视相机，一个相机运动，使目标在水平方向上成像在图像的中心。利用运动相机的转动角度、目标在监视相机的成像坐标以及监视相机的内参数得到目标的三维坐标。这种方法在保证三维测量精度的同时，扩大了感知的视野。实验结果验证了该方法的有效性。再次，给出了仿生眼在运动情况下的深度测量误差模型，分析了减小深度测量误差的方法。并据此给出了基于目标三维坐标的最优观察位姿定义和求取方法。基于最优观察位姿，提出了扫视和平滑追踪的控制方法。在扫视运动中，仿生眼通过立体视觉感知得到目标的三维坐标，根据三维坐标计算仿生眼的最优观察位姿，然后通过运动规划得到步进电机的运行曲线，最后通过运动控制器按照运行曲线控制步进电机到达期望位置。在平滑追踪过程中，测量目标当前时刻的三维坐标，并根据目标的运动速度预测其下一采样时刻的三维坐标，根据预测的目标位置计算最优观察位姿，进而利用步进电机的运动曲线规划，使电机平稳、快速的跟踪目标。实验结果表明，基于三维坐标的扫视和平滑追踪是可行的。最后，以仿生眼为基础，加入可旋转和俯仰的头部，以移动小车为头眼的载体，设计了移动机器人系统平台。利用该平台，提出了一种基于三维坐标的身-头-眼相结合的注视点跟踪方法。在跟踪过程中，利用双目仿生眼得到目标点的三维坐标并分别转换到头部坐标系和移动机器人坐标系。然后根据目标的三维坐标分别得到头部、眼部和躯干的期望位姿。当进行原位注视跟踪时，以资源消耗最少和系统最稳定为原则计算眼部、头部和躯干的期望位姿。当进行趋近注视时，以机器人正对目标且在目标附近为原则计算眼部、头部和躯干的期望位姿。最后控制机器人运动到期望的位姿。实验结果表明，基于三维坐标的身-头-眼相结合的注视点跟踪是可行的。
英文摘要	Bionic eye is an essential part of intelligent robot system. At present, the domestic and foreign researches on bionic eye mainly focus on the bionic motion and visual perception mechanism. However, it is necessary to further study on the combination of the three-dimensional (3D) information of the object and the bionic motion control of the bionic eye. Therefore, in this dissertation, we study the 3D perception of bionic eye and the bionic motion control based on 3D information. The emphases of the dissertation are the 3D perception, saccadic and smooth pursuit control, situ gaze tracking and approaching gaze control of robot. Firstly, a set of 4 degrees of freedom (Dof) bionic eye system is designed. The system acquires images by binocular CCD camera and video capture card. Two cameras are installed on the left and right movement modules. The motion module realizes yaw and pitch by two stepper motors, respectively. The angles of the stepper motors are collected by the rotary encoders. Based on the bionic eye system, this dissertation presents a method of 3D perception based on the monocular motion vision and binocular vision. The transformation relation between the camera coordinate system and the motion module coordinate system is obtained by hand-eye calibration method. The extrinsic parameters of the binocular vision system at any time can be obtained by using the extrinsic parameters of the initial position, the hand-eye parameters of two eyes and the angles of the stepper motors. According to the extrinsic parameters of the binocular vision system, the coordinates of the object in the left and right images and the intrinsic parameters of the cameras, the 3D coordinates of the object can be calculated. According to the extrinsic parameters of monocular vision system at different time, the image coordinates of the object and the intrinsic parameters of the camera, the 3D coordinates of the object can be obtained by the multi-view fusion method of monocular vision system. Through the combination of monocular motion stereo vision and binocular stereo vision, the perceptual field of bionic eye can be enlarged. Secondly, for the bionic eye system, this dissertation proposes a low cost and high precision initial positioning method based on vision and limit switch. When the system is started, the first step is to use the limit switch for coarse positioning, and the second step is the precise vision positioning based on least square ellipse fitting. This method has low cost and high positioning accuracy. When the system starts, the extrinsic parameters calibration of binocular vision system in the initial position do not need repeated. This dissertation presents a stereo vision measurement method based on active object gaze. In the measurement of the 3D coordinates of the object, a camera is fixed, as a surveillance camera. Another camera rotates so that the object is imaged at the center of the image in the horizontal direction. The 3D coordinates of the object are obtained by using the rotation angle of the motion camera, the imaging coordinates of the object in the surveillance camera and the intrinsic parameters of the surveillance camera. This method ensures the accuracy of 3D measurement and improves the perceptual field of vision. Experimental results verify the effectiveness of the proposed method. Thirdly, the depth measurement error model of the bionic eye is given in this dissertation. And the methods to reduce the depth measurement error are analyzed. Then, the definition and the calculation method of the optimal observation position are given. Based on the optimal observation position, the control methods of saccade and smooth pursuit are proposed. The 3D coordinates of the object are obtained by binocular stereo vision. The optimal observation position of the bionic eye is calculated according to the 3D coordinates. Then, the running curves of the stepper motors are obtained by motion planning. The stepper motors are controlled by the motion controller to achieve the desired position. In the process of smooth pursuit, the 3D coordinates of the object at current moment are calculated by binocular stereo vision. According to the moving speed of the object, the 3D coordinates of the object at the next sampling time are predicted. Then the optimal observation position is calculated according to the predicted object position. Lastly, using the motion curves of the stepper motors, the bionic eye can track the object stably and quickly. The experimental results show that the saccade and smooth pursuit based on 3D coordinates are feasible. Finally, a mobile robot system platform is designed based on the bionic eye. The platform includes a 2-Dof head, and a mobile robot as the carrier of the head. Based on the platform, this dissertation presents a new method of body-head-eye gaze tracking based on 3D coordinates. In the process of tracking, the 3D coordinates of the object are obtained by the bionic eye, and then converted to the head coordinate system and the mobile robot coordinate system. The desired position of the head, eyes and body are obtained according to the object’s 3D coordinates. When the situ gaze tracking is performed, the desired positions of the eye, head and body are calculated on the basis of the least resource consumption and the most stable. When approaching the gaze, the robot is expected to face the object in the vicinity of the object. And the desired positions of the eye, head and body are calculated based on this principle. Lastly, the robot is controlled to the desired position. The experimental results show that the body-head-eye gaze tracking based on the 3D coordinates of object is feasible.
关键词	三维感知平滑追踪扫视原位注视趋近注视仿生眼移动机器人视觉测量
文献类型	学位论文
条目标识符	http://ir.ia.ac.cn/handle/173211/14718
专题	毕业生_博士学位论文
作者单位	中国科学院自动化研究所
第一作者单位	中国科学院自动化研究所
推荐引用方式 GB/T 7714	王庆滨. 仿生眼三维感知与运动控制研究[D]. 北京. 中国科学院研究生院,2017.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
王庆滨博士论文全文.pdf（4838KB）	学位论文		限制开放	CC BY-NC-SA