中文摘要：

近年来国外发展起来的大尺度粒子图像测速方法是一种针对野外环境非接触式河流水面流速测量的有效手段.然而由于测量中播撒的人工示踪粒子极易受到水面光学噪声的干扰,往往导致观测目标的检测和运动矢量估计出现较大误差,最终造成流量测验失败.针对传统基于可见光光强的成像装置在获取水面图像时存在对比度低、目标信息丢失严重等问题,基于偏振成像探测技术,提出了一种三通道CMOS（图像传感器）同步偏振成像及示踪粒子（观测目标）检测方法.设计了三通道CMOS同步偏振成像装置,可同步采集3个不同方向的水面偏振图像,从而计算线偏振度图像.针对3路CMOS采集的偏振图像之间存在像元不对齐,导致计算线偏振度图像时出现虚假目标的情况,提出了一种改进的SIFT特征匹配算法,可实现三幅偏振图像的精确配准,提高了线偏振度图像的求解精度.实验表明,与传统的光强成像方法相比,本方法能稳定可靠地检测出水面示踪粒子,目标检测率由58.8％提高到88.2％,为大尺度粒子图像测速提供了可靠有效的水面观测目标检测方法.

英文摘要：

large-scale particle image velocimetry （LSPIV） developed abroad recently is an effective method for the non-contact measurement of the river velocity in field environment. However, the tracer particles sowed in the meas- urement are subject to the disturbance causing by the water-surface optical noise, which leads to the deviation in tar- get detection and motion vector estimation, and i~nally causes the measurement failure. Aiming at the issues that the water surface images acquired by traditional image acquisition devices based on visible light often suffer the serious problems,such as low-contrast and target information losing, this paper proposes a method for the detection of the tracer particles based on the polarization imaging detection with 3 channel CMOS image sensor. A three-channel CMOS synchronous polarization imaging system was designed to acquire polarization images in three directions syn- chronously, and then the images of degree of linear polarization can be calculated. Aiming at the pixel alignment prob- lem between successive images that leads to fake targets when calculating the images of degree of linear polarization （DoLP）, an improved SIFF feature matching algorithm is proposed to perform exact image registration of the three polarization images and improve the accuracy of the images of DoLP. The experiment results show that compared with traditional imaging method based on visible light ,this method can stably and reliably detect the water-surface tracer particles ,and the target detection rate is improved from 58.8% to 88.2% ;thus the proposed method is an effectivemethod of target detection for LSPIV.