运动重建约束角的圆位姿二义性消除方法  

Solution of Duality in Pose Estimation of a Single Circle Utilizing Constraint Angles Motion Reconstruction

作　　者：张李俊[1] 黄学祥[1] 冯渭春 胡天健[1] 梁书立[1] 

机构地区：[1]北京跟踪与通信技术研究所,北京100094

出　　处：《光学学报》2016年第1期187-196,共10页Acta Optica Sinica

基　　金：国家自然科学基金(11402004)

摘　　要：针对单个圆进行视觉测量存在的位姿二义性问题,提出了一种运动重建约束角的二义性消除方法。在相机已标定的前提下,利用圆的投影椭圆曲线确定圆的位姿参数,位姿存在二义性;利用多视图从运动恢复结构方法估计特征点的世界坐标,并利用特征点重建约束角;基于刚体运动不改变空间夹角的特性同时唯一确定多视图的圆位姿。仿真结果显示:在500 mm距离内,噪声强度为0.4 pixel时二义性消除成功率为100%,强度为1 pixel时成功率约为93%;噪声强度为0.1 pixel,距离在1000 mm时具有99.6%的成功率。此外,在二视图的情形下算法的平均运行时间约为5ms,适合实时性的应用。该方法不需要目标的额外信息,能够实时有效地消除圆位姿的二义性。In order to cope with the pose duality problem in the visual measurement of a single circle, a disambiguating approach is proposed utilizing constraint angles motion reconstruction. On the premise of camera calibrated, the pose parameter of a circle is estimated from its projected ellipse curve but the pose has duality. The world coordinates of feature points are recovered by the method of multiple view structure from motion and constraint angels are reconstructed with feature points. Based on the nature that rigid body motion does not change space angles, the poses of multiple view are determined simultaneously. Experimental results indicate that within the range of 500 mm the circle radius, the success rate of eliminating false solution is up to 100% when the noise level is within 0.4 pixel and 93% when the noise level is 1 pixeL And the success rate is 99.6% when the noise level is 0.1 pixel and the range is 1000 mm the circle radius. Besides, the average run time of the proposed algorithm in two views is about 5 ms which suits real-time application. The proposed approach can effectively identify real poses of a single circle in real time without any additional information of target.

关 键 词：机器视觉 圆 位姿估计 二义性 角度约束 从运动恢复结构 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]