A solution for camera tracking based on coplanar circles from industrial objects was presented. Firstly, a coordinate system for an object frame was constructed based on coplanar circles. Then, the Extended Kalman Filter(EKF)was exploited to predict camera pose to restrict the positions and regions for searching ellipse contours, and the candidate contour points of ellipse were formed by calculating normalized cross coefficient of gradient. With maximum likelihood estimation, a high accurate ellipse was obtained after eliminating outliers from candidate dataset. Finally,the camera pose was estimated using projective theory of conies, and the filter state was updated. Experiments show that the maximum pose deviation of camera tracking is 1. 4° for rotation and 3. 5 mm for translation, and the tracking rate is 10～12 frame/s. The results can meet the requirement of pose tracking in industrial Augmented Reality(AR).%提出了利用工业对象表面共面圆进行相机位姿跟踪的解决方案.首先,利用共面圆构建对象坐标系；用扩展卡尔曼滤波器预测相机位姿限定椭圆轮廓点搜索的位置及范围,计算梯度归一化互相关系数来形成轮廓点候选数据集；然后,剔除候选数据集中的野值,用最大似然估计法进行高精度椭圆拟合；最后,根据二次曲线的射影理论计算相机位姿并更新滤波器的状态.实验表明,本文方法跟踪相机位姿的最大误差角度为1.4°,平移为3.5 mm,跟踪速度为10～12frame/s,满足工业增强现实中相机位姿跟踪的应用需求.
展开▼
