直升机桨叶图像中圆形标记点圆心检测及修正方法

视觉测量直升机高速旋转桨叶形变常采用基于圆形标记点的测量方法，但该方法中圆形标记点具有低曝光、小目标和投影不对称等特点，极易产生标记点漏检和圆心坐标误差的问题。为减小漏检、修正圆心坐标误差，提出了一种直升机桨叶图像中圆形标记点圆心检测及修正方法。首先，提取图像中局部极值中心的像素坐标，并依据阵列排布结构滤除干扰，获得所有圆形标记点极值中心的像素坐标；其次，以各极值中心的像素坐标为圆心，与相邻极值的最小距离为直径，建立圆形ROI （region of interest）区域，在ROI区域内并行分水岭变换和最小二乘法圆拟合得到圆心；再次，采用透视变换建立该图像与垂直相机光轴的同相位桨叶图像（正面图像）的投影映射关系，再采用LM （levenberg-marquardt）优化求解投影映射矩阵；最后，将该图像转换为正面图像进行圆心检测，再将该圆心坐标逆变换得到精确圆心坐标。实验结果表明，本文检测方法准确率和精度分别达98.89%和0.191 mm，已应用于直升机高速旋转桨叶运动轨迹和形变的高精度视觉测量。

Circle Center Detection and Correction Method of Circular Markers in Helicopter Blade Image

The deformation measurement of rotating helicopter high-speed rotor blades is usually based on circular markers. However, circular markers are usually with problems, such as low exposure, small target region, and asymmetric projection, which easily lead to detection missing of markers and location error of circle centers. In order to avoid these problems, a detection method based on circular markers in helicopter rotor blade images is proposed in this paper. Firstly, pixel coordinates of local gray extreme value centers in the image are extracted, and interferences are removed according to array arrangement structure, then pixel coordinates of all circular marker extreme values are obtained. Secondly, the circular region of interest (ROI) is established with each extremum coordinate as the center and the minimum distance from the adjacent extremum as the diameter. Within the ROI, the circle center is obtained by parallel watershed transformation and least square circle fitting. Thirdly, based on perspective transformation, the projection mapping relationship between the image and another helicopter rotor blade image, which with the same phase and perpendicular to the camera optical axis is established. And the projection mapping matrix is optimized by levenberg-marquardt (LM). Finally, the images is converted to a positive image for circle center fitting, and exact circle center coordinates are obtained by inverse transformation of circle center coordinates. Experimental results show that the accuracy and the precision of the proposed method are 98.89% and 0.191 mm, and it has been applied in the high-precision visual measurement of the motion trajectory and deformation of the helicopter high-speed rotor baled.