基于对偶四元数的单目视觉目标位姿测量

目的在机器人视觉应用领域中,为控制机器人能够完成焊接、搬运、跟踪等任务,需要确定摄像机与目标之间的相对位姿关系,提出一种目标位姿测量方法。方法利用单摄像机获取目标特征,坐标变换参数表示为对偶四元数的形式,同时计算旋转矩阵和平移向量,构建位置向量和方向向量的测量值与模型值之间的误差方程,利用Hopfield神经网络实现拉格朗日乘子法,求解目标位姿最优解。结果利用Matlab软件平台,选择SVD,DQ以及文中算法进行比较,仿真实验结果表明,基于Hopfield神经网络和对偶四元数的位姿测量算法计算出的位姿参数误差最小。随着测量点数量的增大,文中提出的算法精度更高。结论对偶四元数同时求解位姿变换矩阵的旋转分量和平移分量,可消除计算误差,基于Hopfield神经网络和拉格朗日乘子法,可快速准确地计算,并收敛至目标位姿最优解。

Monocular Vision Measurement of Object Pose Based on Dual Quaternion

The work aims to propose a method to measure target pose regarding the relative pose relationship between the camera and target to be determined to control robot to complete wielding, transporting, tracking, etc. in the application field of robot vision. Target characteristics were acquired with single camera and the coordinate transformation parameters were expressed as dual quaternion. Meanwhile, rotation matrix and translation vector were calculated. The error equation between measured values and model values of position vector and direction vector was built. Hopfield neural network and Lagrange multiplier method were used to solve the optimal solution of target pose. Through Matlab software platform, SVD, DQ and the proposed algorithm were selected for comparison. The simulation test results showed that the error of pose parameters calculated by the pose measurement algorithm based on Hopfield neutral network and dual quaterion was the minimum. With the increase in the number of measurement points, the proposed algorithm was of higher accuracy. The simultaneous solution to rotation and translation components of pose transformation matrix for dual quaternion can eliminate calculation errors. Hopfield neural network and Lagrange multiplier method can calculate and converge to the optimal solution of target pose quickly and accurately.