基于Broyden在线图像雅可比矩阵辨识的视觉伺服

Vision servoing based on online estimation of image Jacobian matrix of Broyden

为了在显微视觉中进行无标定的视觉伺服任务,提出了一种基于切比雪夫多项式构成成本函数的Broyden图像雅可比矩阵估计方法.比较了由递归最小二乘算法构成成本函数和由切比雪夫多项式算法构成成本函数的特点,在不依赖先验知识的情况下,切比雪夫多项式算法构成成本函数的Broyden图像雅可比矩阵估计方法有较好的收敛速度和系统性能.对多个微小目标物体和末端执行器应用了模糊C均值聚类进行分类与识别,然后根据得到的图像雅可比矩阵辨识器,在显微视觉环境下进行了微小物体的跟踪实验,仿真和实验验证了算法的有效性和可行性.

A Broyden method with Chebyshev polynomial as a cost function is presented to estimate image Jacobian matrix in the uncalibrated microscope vision servoing. Compared with recursive least square algorithm which is used to construct the cost function, Chebyshev polynomial algorithm without the prior knowledge has also the great adaptability on convergence speed and stability. Fuzzy C- mean cluster to recognize and classify objects and end-effectors was used. Location and tracking tests of micro objects were presented based on image Jacobian model we developed. The performance was confirmed by simulations and experiments.