【作者】 胡涛；

【导师】 李丹勋；

【作者基本信息】 清华大学 ， 水利工程， 2010， 硕士

【摘要】 随着计算机和图像处理技术的发展,流场测速实现了从单点测量到全场测量的飞跃。目前应用图像处理技术进行流场测量,主要有PIV和PTV两种方法。其中PTV通过直接跟踪流场中示踪粒子的运动轨迹而求得速度,准确直观,而且避免了PIV中出现的平均效应,在测量某些复杂流场方面具有明显优势,所以不断得到发展和应用。PTV技术主要包括粒子识别与粒子匹配两个过程,而粒子匹配过程中使用的匹配算法一直是PTV技术的研究重点之一。本文对比分析了六种不同的匹配算法,包括最近邻法、四帧图像粒子跟踪算法、PCSS算法、弹簧模型算法(SPG)、匹配几率法和速度梯度张量算法。为评价种PTV匹配算法的优劣,主要从两个方面进行考虑:运算效率和算法精度,即算法运算所需的时间长短和算法处理结果的准确性高低。为了定性对比各种算法的运算效率,引入了算法的“时间复杂度”这一概念,用每种算法包含的基本运算操作的次数作为评价算法用时的标准。基本操作多数情况下是由算法最深层环内的语句表示的。为了定量对比算法的特点,定义了匹配率Φ_r和误匹配率Φ_e两个参数,通过比较算法处理图像所需的时间以及处理结果的误匹配率大小,就可以定量描述算法的运算效率和计算精度。使用三种不同的人工模拟规则流场的图像序列,对算法的适用范围进行检测,包括水平不等速流动、波浪型不等速流动和漩涡型不等速流动,其中每种类型的流场包含三组粒子密度不同的图片序列。应用不同的匹配算法处理同一组图像序列,对比算法的处理结果,分析各算法在处理该类型流场时的优缺点,同时对比同一算法处理不同流场图像序列时在运算效率和精度上产生的差异,总结出该算法的适用性。研究结果表明,在所对比的六种匹配算法中,匹配几率法处理规则流场图像的效果最好,在运算效率和精度上都能满足实际应用的需求,最近邻法的运算效率最高,在粒子密度较低且流场流速较小时精度也很好,PCSS算法精度较好,但运算效率不高,而其它三种算法的精度相对都比较差。更多还原

【Abstract】 Velocity measurement of fluid flows in an entire area rather than at a single point has been achieved thanks to the fast development in computer science and image processing technique. At present, two image-processing-based velocity measuring methods are widely used, namely, PIV (Particle Image Velocimetry) and PTV (Particle Tracking Velocimetry). Compared with PIV, PTV provides more accurate velocity information, for it directly tracks the motion of tracer particles seeded in the flow and stays free from the averaging effect inherent in PIV. Due to its advantages in measuring certain complex flows, PTV has been experiencing fast development aimed at wider applications.PTV has two major sub-processes, namely, particle identification and particle matching. Algorithm for particles matching, essential for the success of PTV, has been the focus of PTV research. This paper made comparisons of six PTV matching algorithms, including the nearest-neighbor algorithm, the four-frame tracking algorithm, the PCSS algorithm, the SPG algorithm, the matching probability algorithm, and the velocity-gradient tensor algorithm.Comparison of various algorithms was made in terms of calculating efficiency and matching accuracy. A parameter called“time complexity”for the algorithms, defined as the times of basic process being repeated, was proposed to indicate the efficiency.Two other parameters, the matching rateΦ_rand false matching rateΦ_e , were introduced for quantitative comparison of these algorithms in terms of efficiency and accuracy. The comparison was based on three categories of artificial image sequences of various flows, including a horizontal flow, a wavy flow, and a vortex flow, and each category has images of various particle densities. Comparison of the same algorithm was made by using different image series, and then comparison of various algorithms was made by using the same image series.The results show that: (1) the matching-probability algorithm provides the best results in terms of accuracy, (2) the nearest neighbor method is most efficient under low particle density and small flow velocity, (3) the accuracy of the PCSS algorithm is good, but its efficiency is low, and (4) the accuracies of the other three algorithms are all very poor.更多还原