串联机器人实时双目视觉定位及跟踪技术研究

发布时间：2018-05-28 03:00

本文选题：双目视觉 + 视觉伺服　；参考：《湖北工业大学》2017年硕士论文

【摘要】：基于双目视觉的串联机器人视觉伺服系统能够直接获取目标的位置信息和更多的图像特征,实现了基于视觉的闭环控制,是视觉伺服的典型结构。本文围绕着实时性、稳定性、准确性将课题分为目标在图像中的定位、目标在三维中的定位、视觉伺服控制方法、目标位置预测和跟踪四个部分分析,分别研究其涉及的基本理论和方法,并通过仿真或理论评估这些方法对课题的适用性,探究串联机器人实时双目视觉伺服稳定有效的实现方法。(1)使用目标在图像上的点特征定位目标,并通过对常用点特征提取算法原理和特点的对比,了解到Ransac优化Orb算法具有强实时性和稳定性。(2)分析了双目下的目标由图像空间到三维空间的映射过程,发现视差的准确性影响着三维定位的准确性。引出了比传统立体匹配算法更优的ELAS立体匹配算法,并针对该算法误匹配率较高的问题,提出了基于视差连续性约束的改进ELAS立体匹配算法,至少降低了原算法15.63%的误匹配率。(3)基于eye-in-hand配置的6自由度串联机器人和静止的点目标,对比了PBVS、IBVS、HBVS的视觉伺服方法。针对现有IBVS中伪逆构造法的缺点,提出了基于混合的雅可比伪逆构造法,仿真验证了新构造法的有效性。(4)针对运动目标轨迹预测的问题,讨论了已知目标运动状态下的线性kalman滤波、已知目标运动状态下多传感融合的扩展kalman滤波以及未知目标运动状态下的交互多模型的kalman滤波对目标的跟踪预测,并仿真验证了这些方法在图像和三维跟踪过程中的有效性。在实验中,跟踪平稳误差均小于5个像素或5mm。最终获得了串联机器人实时双目视觉伺服稳定有效的实现方法,即使用Ransac-Orb提取目标点特征,使用改进的ELAS求取视差,使用基于基于混合的雅可比伪逆构造法的IBVS作为视觉伺服控制律,使用对应场景下的kalman滤波对运动目标进行轨迹预测。
[Abstract]:The visual servo system of series robot based on binocular vision can directly obtain the position information of the target and more image features. It realizes the closed-loop control based on vision and is a typical structure of visual servo. This thesis is divided into four parts: target location in image, target positioning in 3D, visual servo control method, target position prediction and tracking, around real-time, stability and accuracy. The basic theories and methods involved are studied, and the applicability of these methods to the subject is evaluated by simulation or theory. This paper probes into the stable and effective realization method of real-time binocular visual servo of serial robot. It uses the point feature of the target to locate the target on the image, and compares the principle and characteristics of the common point feature extraction algorithm. It is found that the Ransac optimized Orb algorithm has strong real-time and stability. (2) the mapping process from image space to 3D space is analyzed. It is found that the accuracy of parallax affects the accuracy of 3D location. The ELAS stereo matching algorithm is better than the traditional stereo matching algorithm. Aiming at the problem of high mismatch rate, an improved ELAS stereo matching algorithm based on parallax continuity constraint is proposed. At least the mismatch rate of the original algorithm is reduced by 15.63%.) based on the eye-in-hand configuration, the 6-DOF series robot and the stationary point target are compared. The visual servo method of the PBVS IBVS / HBVS is compared. In view of the shortcomings of the existing pseudo inverse construction method in IBVS, a mixed Jacobian pseudo inverse construction method is proposed. The simulation results show that the new construction method is effective to predict the trajectory of moving targets. In this paper, the linear kalman filter with known target motion, the extended kalman filter with multi-sensor fusion and the target tracking prediction with interactive multi-model kalman filter under unknown moving state are discussed. Simulation results show that these methods are effective in image and 3D tracking. In the experiment, the tracking stationary error is less than 5 pixels or 5 mm. Finally, a stable and effective realization method of real-time binocular visual servo of serial robot is obtained, that is, the feature of target point is extracted by Ransac-Orb, and parallax is obtained by using improved ELAS. The IBVS based on the mixed Jacobian pseudo-inverse method is used as the visual servo control law, and the kalman filter in the corresponding scene is used to predict the trajectory of the moving target.
【学位授予单位】：湖北工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【参考文献】