双目视觉手眼标定的精度优化

Research on the Optimization Algorithm of Hand Eye Calibration Accuracy in Binocular Vision

双目视觉定位是机器人领域对环境感知的常用技术,手眼标定的好坏很大程度上影响着机器人定位精度。传统的线性标定只需一组数据就可标定,但极易受到偶然误差的影响;非线性算法使用多组数据,但容易出现奇异值和陷入局部最优点。针对这两个问题,用非线性算法Ada-delta对线性算法Tsai得到的数据进行优化,非线性算法多组数据运算避免了偶然误差,线性求解的结果充当优化算法的常数项也避免了出现奇异值的情况。用优化前后的手眼标定矩阵求得理论机械臂末端位姿,分别与实际位姿作差,取其Frobenius范数。结果证明:优化后的Frobenius范数为0.30565小于优化前的0.520717,优化效果良好。

Binocular vision positioning is a common technology in the field of robot environment perception.The hand eye calibration has a great impact on the positioning accuracy of robot.Traditional linear calibration only needs a set of data to calibrate,but it is easily affected by accidental errors;Nonlinear algorithms use multiple sets of data,but they are prone to singular values and local optima.To solve these two problems,the nonlinear algorithm Ada-delta is used to optimize the data obtained by the linear algorithm Tsai.The multi group data operation of the nonlinear algorithm avoids accidental errors,and the results of the linear solution act as the constant term of the optimization algorithm to avoid the occurrence of singular values.Using the hand eye calibration matrix before and after optimization,the end pose of the theoretical manipulator is obtained,and the difference is made with the actual pose,and the Frobenius norm is taken.The results show that the Frobenius norm after optimization is 0.30565,which is less than 0.520717 before optimization,and the optimization effect is good.