Abstract: An end-face butt-joint assembly has important applications in the aerospace, automatic assembly, human-computer interaction, and other fields.However, there are still some problems with butt-joint detection, such as a non-unique circular attitude solution, low solution accuracy, and low docking efficiency.To improve the measurement accuracy and robustness, taking the back against vision and the arrangement of the point lasers as the means of measurement, an end face butt-joint positioning method based on a normal constraint is proposed.First, images of the circular features of a butt-joint were collected using back against vision, and based on the edge extraction of the circle features, the Random Sample Consensus (RANSAC) algorithm is used to fit the circular features, reduce the influence of image noise, and improve the robustness of the feature fitting.Second, according to the model of a circular pose solution using a monocular camera, the ambiguous center position of the circle and the pose solution are obtained.Third, the plane normal vector obtained by fitting the arranged laser points was taken as the normal constraint, and when the angle between the circular attitude solution and the direction constraint is the smallest, the circular attitude solution is taken as the correct one.Finally, to complete the butt-joint assembly, a robot conducted the pose and displacement adjustments.The experiments using a robot with a small angle pose and a small displacement show that the proposed method achieves a high pose adjustment accuracy and a fast convergence of the displacement adjustment.The maximum deviation of the pose angle is within 0.03°, and the maximum deviations of the center displacement in the X- and Y- directions are 0.01 and 0.03 mm, respectively.

Key words: back against vision, normal constraint, butt-joint assembly, circular pose, Random Sample Consensus(RANSAC) algorithm

摘要： 端面对接在航空航天、自动化装配、人机交互等领域有着重要应用，然而端面检测仍存在圆姿态解不唯一、解精度不高、对接效率较低等问题。为提高测量精度和鲁棒性，以背靠视觉和排列点激光为测量手段，提出一种基于法向约束的端面对接定位方法。通过背靠视觉分别采集端面圆特征图像，在对圆特征进行边缘提取的基础上采用随机采样一致性算法进行圆特征拟合，降低图像噪声的影响，提高特征拟合的鲁棒性。根据单目相机圆位姿解算模型，得到二义性的端面圆中心位置和姿态解，将排列点激光拟合得到的平面法向量作为法向约束，并取圆姿态解与法向约束的夹角最小者为正确的圆姿态解。最后，由机器人先后进行姿态和位移迭代调整，完成端面的对接装配。机器人小角度姿态、小位移的实验结果表明，该方法具有姿态调整准确度高、位移调整收敛快的特点，姿态最大偏差为0.03°，中心位移X方向最大偏差为0.01 mm、Y方向最大偏差为0.03 mm。

关键词: 背靠视觉, 法向约束, 对接定位, 圆位姿, 随机采样一致性算法

CLC Number: 

• TP391.41