Abstract: For the peak signals such as spectrogram, electrocardiogram, light spot image of linear CCD, it is difficult to denoise the image completely and preserve the peaks’ features effectively during filtering. The location, amplitude, area, and edge steepness of the signal peak are inevitably affected. Based on the analysis of conventional Weighted Moving Average（WMA） filtering algorithm, a new bivariate distance weight concept based on amplitude & phase is proposed. It takes into account not only the phase correlation but also the amplitude correlation among the sampled pixels during WMA computation, and solves the problem pushing up the data on the bottom of the peak. By dynamically adjusting one or both of the phase distance threshold and the amplitude distance threshold, the problem pulling down the data on the top of the peak is also solved. Finally, an adaptive moving average filtering algorithm weighting on amplitude & phase distance is developed. This new algorithm is applied to the vision testing system of vehicle wheel alignment parameters based on the linear CCD image, and improves the sub-pixel measurement precision of the light spot position from 0.5 pixels to 0.03 pixels.

Key words: adaptive filtering, weighted moving average, phase distance, amplitude distance

摘要： 对于光谱图、心电图、线阵CCD光斑图等尖峰型信号，既要滤除噪声又要保持信号峰的位置、幅度、面积、边缘陡度等特征，这是一对难以兼顾的矛盾。在分析了常规加权滑动平均滤波算法的基础之上，提出了幅相二元距离权系数的概念。它不仅考虑了参与加权滑动平均运算的采样数据点间在相位上的相关性，同时也考虑了它们在幅值方面的相关性，有效地解决了峰脚数据被拉高的问题。通过动态调整相位距离阈值或幅值距离阈值，有效地解决了峰顶数据被拉低的问题，从而建立了基于幅相二元距离加权的自适应滑动平均滤波算法。该算法已成功应用于基于线阵图像的视觉式全参数车轮定位检测系统中，使得光斑位置亚像素检测精度从原来的0.5像素提高到了0.03像素。

关键词: 自适应滤波, 加权滑动平均, 相位距离, 幅值距离