针对波纹板 － 箱梁 T 型接头角焊缝位置的激光双目视觉检测，提出了一种变步长角点检测算法，即先建立波纹板角焊缝位置的双目视觉测量模型，根据激光条纹投射在角焊缝上的成像特征进行图像预处理，然后设计变步长搜索算法先找到激光条纹上的近角点，于近角点两侧选取激光条纹上的像素点构造两条直线，由直线的交点来确定角点．试验结果表明，用该方法测量角焊缝位置坐标，可靠性高，测量误差范围在 0. 1 mm 以内，角焊缝轨迹三维重建误差在 1mm 以内，测量精度优于传统算法

Proposed in this paper is a variable- step corner detection algorithm for the laser binocular vision detec-tion of fillet weld position of the T- type joint between corrugated plate and box girder.In this algorithm,a binocularvision detection model of fillet weld position is established and an image preprocessing is performed for the laserstripe images according to the imaging characteristics of the stripe projected on the fillet weld.Then,a variable-step searching algorithm is developed to find the near- corner point in the laser stripe,and several pixels at bothsides of the near- corner point are selected to construct two straight lines.Finally,the corner point of the fillet weldis determined according to the intersection point of the two constructed lines.Experimental results show that theproposed algorithm is superior to those traditional ones due to its high reliability,low detection error (less than0. 1mm) and low 3D trajectory reconstruction error of fillet weld (less than 1mm).