频域时空图像测速法的图像滤波器敏感性分析

时空图像测速法是以河流表面图像中测速线为分析区域、通过检测合成时空图像的纹理主方向计算得到一维时均流速的测量方法，具有空间分辨率高、实时性强的特点。在实际应用中纹理主方向的检测精度难免会受到水面紊流、倒影、耀光、障碍物、降雨等环境扰动的影响，导致测量出现粗大误差。频域滤波技术是一种抑制噪声的有效方法，能显著提高时空图像的纹理清晰度。但现有研究在滤波器参数的敏感性分析方面存在不足，使得该方法的适用性受限。对此通过在水文站搭建在线视频测流系统采集了不同条件的河流水面视频数据，分析了6种典型场景下时空图像的空域及频域特性，进而开展了频域扇形滤波器方向角、通带夹角及半径参数的敏感性分析实验。实验结果表明：采用提出的椭圆形积分区域检测方向角优于现有的单像素宽直线；当设置通带夹角为±5.3°且半径为R/2时，滤波器在上述场景下均能有效地滤除噪声干扰。使得时空图像纹理主方向的检测精度在正常场景下达到0.1°,在复杂含噪场景下控制在0.5°以内，表面流速测量的相对误差小于6.2%。

Sensitivity analysis of image filter for space-time image velocimetry in frequency domain

The space-time image velocimetry (STIV) is a time-averaged velocity measurement method,which takes a testing line as the interrogation area (IA), and detects the main orientation of texture (MOT) of a generated space-time image ( STI) to estimate the 1D velocity. It has the characteristics of high spatial resolution and real-time. In practical applications, the detection accuracy of MOT is inevitably affected by environmental disturbances such as turbulence, shadow, flare, obstacle, and rain on river surface, which result in gross errors in the measurement. The image filtering technology in frequency domain is an effective method for restraining noise, which can significantly improve the texture clarity of STIs. However, the existing researches are insufficient in the sensitivity analysis of filter parameters, which limit the applicability of this method. In view of this, the video data of river surface under different conditions are collected by establishing an online video-based flow measurement system at a hydrological station. The spatial and frequency domain characteristics of STIs in six typical scenarios are analyzed. The sensitivity of three parameters of a fan filter in frequency domain, including direction angle, passband angle and radius is determined. Experimental results show that the proposed ellipse integral region is better than the existing single-pixel-wide line to detect direction angle. When the passband angle is set to be ±5. 3° and the radius is R/ 2, the filter can effectively filter out the noise interference in the above scenarios. The detection accuracy of MOT reaches 0. 1° in the normal scene and is controlled within 0. 5° in the complex noisy scene. The relative error of the surface flow velocity measurement is less than 6. 2% .