针对航空密封胶与预装配紧固件等因素对自动化装配系统基准检测的干扰问题以及系统对多类型装配基准检测的需求，分析了工业环境装配基准特点以及目前检测算法的局限性，提出了一种基于单目视觉的多类型装配基准椭圆特征稳健检测方法。该方法以弓弦比与弦长为依据选取最优弧，通过分析剩余弧段与最优弧的关联凸性，实现对弧段的快速筛选。借助改进的基于存在概率的圆检测算法拟合装配基准近似圆，并通过设定的距离阈值实现对圆弧的快速精确聚类。最后，采用直接最小二乘法拟合装配基准椭圆轮廓，并通过参数约束对拟合结果进行去伪。通过现场测试与精度验证，算法对装配场景常见干扰因素具有显著的抑制作用，对基准孔的检测精度（孔径）与定位精度（孔距）分别为0.10 mm与0.09 mm，对铆钉、靶标点、通孔以及含穿心夹基准孔的检测准确率为97.9%、98.3%、99.1%和91.1%，能够满足自动化装配系统对多类型装配基准的检测需求。

Regarding the disturbance problem of automatic assembly system reference hole detection due to such factors like aviation sealant and tack fasteners， this paper analyzes the characteristics of assembly reference hole in industrial environment and the limitations of current detection algorithm， and proposes a robust detection method of multi-type assembly reference hole based on monocular vision. This method selects the optimal arc based on the bow-string ratio and chord length. By analyzing the associated convexity between the optimal arc and residual arcs， the rapid filtering of arc segments is realized. An improved circle detection algorithm based on the probability of existence map is proposed to fit the approximate circles of assembly reference hole. The accurate clustering of arc segments is achieved by the set distance threshold and the approximate circle of assembly reference hole. Finally， ellipse fitting is performed by using direct least-squares method. And the false alarms are subsequently removed. Through field testing and accuracy verification， the algorithm has a significant inhibitory effect on the common disturbance factors in the assembly scene. The detection accuracy （aperture） and positioning accuracy （hole spacing） of the reference hole are 0.10 mm and 0.09 mm respectively. The average recall rates of the algorithm for the detection of multi-type assembly reference hole of rivets， target points， through-holes and holes with piercing clamps are 97.9%， 98.3%， 99.1% and 91.1% respectively， which can meet the detection requirements of automatic assembly system.