Abstract: The diameter of the conical surface detection which is the calibration line of the valve dimensions is the important measurement in automotive engine valve size detect. To solve application of Prewitt operator for bevel detection algorithm detects exists slow, poor stability and low precision in high-precision measurement（accuracy in μ level）, an improved detection algorithm for the diameter of the hypotenuse is presented. The algorithm median filtering the image, and then locate the rough edge of the target area to determine the direction for the second edge detection, and use quadratic curve fitting algorithm to get precise sub-pixel coordinates. Multiply the difference between the left and right edge of the coordinates by pixel equivalent and equivalent to error compensation to get precise diameter, and finally find the diameter gauge position. Experimental results show that the algorithm with existing algorithms, the accuracy increased by 1 times, stability and computational speed has improved greatly.

Key words: conical surface, Prewitt operator, sub-pixel, curve fitting, edge detection, edge locate

摘要： 圆锥面量规直径的检测是汽车发动机气门检测中重要的尺寸测量，是气门各项长度尺寸检测的基准线。针对Prewitt算子在边缘检测上存在的速度慢、稳定性差以及在高精度（精度在μ级）测量中准确度低等问题，提出了一种改进的斜向边缘直径检测算法。对图像进行中值滤波，对目标区域边缘进行粗略定位以确定第二次边缘检测的方向。在确定的检测方向上，应用二次曲线拟合算法得出精确的亚像素边缘信息，将测得的左右边缘坐标差乘上像素当量并进行误差补偿，得到精确的直径值，寻找量规直径位置。实验结果表明，算法与已有算法相比，准确性提高1倍，稳定性与运算速度有较大提高。

关键词: 圆锥面, Prewitt算子, 亚像素, 曲线拟合, 边缘检测, 边缘定位