帧差与快速密集光流结合的图像法测流研究

图像法测流技术因其简便、高效、安全等优点得到了普遍的关注,开始应用于国内外水文站。目前主流的图像法测流技术通常采用天然水面模式作为示踪物,利用空域互相关法处理图像获得流场矢量,被称为大尺度粒子图像测速(LSPIV)。但该方法在天然河道流量测量中仍存在着较大不确定性及一致性差等问题,并且由于相关性匹配问题该方法的测量速度很难达到实时性测量要求。本文提出一种基于帧间差分与快速密集光流结合的分组图像法(FD-DIS-G)快速测流方法。首先,利用帧差法计算运动显著性图去捕捉到细微的水面运动来处理低流速下河流运动在视频中表现不明显的问题。其次,使用快速密集光流法(DIS)计算运动显著性图中小块区域之间的密集光流,小块区域和快速密集光流法结合能提高流量测量的精度和时效性。同时,设计了一种新的分组处理奇异值的方法,提高了算法的整体准确性,增强了算法的稳定性。本文将流速仪测量得到的垂线平均流速、平均流速以及断面流量作为比测标准,利用水文站所拍得的天然河道水流视频进行比测实验,实验结果表明,相比于广泛使用的LSPIV测量方法,本文方法在平均流速和断面流量上的精度有明显的提升,垂线平均流速的稳定性有显著的增强,并且实时性好。

Image Flow Measurement Based on the Combination of Frame Difference and Fast and Dense Optical Flow

Image flow measurement technology has received widespread attention due to its simplicity, efficiency, and safety, and has begun to be applied to hydrological stations at home and abroad. The current mainstream image flow measurement technology usually uses the natural water surface pattern as a tracer and uses the spatial cross-correlation method to process the image to obtain the flow field vector, which is called large-scale particle image velocimetry (LSPIV). However, this method still has problems such as large uncertainty and poor consistency in natural river flow measurement, and the measurement speed of this method is difficult to meet the requirements of real-time measurement owing to the correlation matching problem. This paper proposes a fast flow measurement method based on the combination of frame difference and fast dense optical flow (FD-DIS-G). Firstly, calculating the motion saliency map to capture the subtle water surface motion by using the frame difference method to deal with the problem that the river motion at low flow velocity is not obvious in the video. Secondly, the fast dense optical flow method (DIS) is used to calculate the dense optical flow between small areas in the motion saliency map. The combination of small areas and the fast dense optical flow method can improve the accuracy and timeliness of flow measurement. At the same time, a new method of grouping processing abnormal values is designed, which improves the overall accuracy of the algorithm and enhances the stability of the algorithm. In this paper, the average vertical velocity, average velocity, and cross-sectional flow measured by the flow meter are used as the comparison standards, and the comparison experiment is carried out using the natural river flow video taken by the hydrological station. The experimental results show that compared with the widely used LSPIV measurement method, the method in this paper has a significant improvement in the accuracy of the average flow velocity and cross-sectional flow, and the stability of the vertical average flow velocity is significantly enhanced, and the real-time performance is good.