The application of a 3D PTV algorithm to a mixed convection flow
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Experiments in Fluids 33 (2002) 603–611 DOI 10.1007/s00348-002-0513-9 The application of a 3D PTV algorithm to a mixed convection flow R.N. Kieft, K.R.A.M. Schreel, G.A.J. van der Plas, C.C.M. Rindt 603 Abstract A 3D particle-tracking velocimetry (PTV) algo- PTV techniques, individual particles in subsequent images rithm is applied to the wake flow behind a heated cylinder. are tracked, whereas in PIV techniques the average dis- The method is tested in advance with respect to its accu- placement is determined within a segment of an image racy and performance. In the accuracy tests, its capability (interrogation area). to locate particles in 3D space is tested. It appears that the In 2D techniques, the flow is illuminated with a thin algorithm can determine the particle position with an ac- light sheet and only the velocity components within this curacy of less than 0.5 camera pixels, equivalent to sheet can be evaluated. Although a few methods exist for 0.3 mm in the present test situation. The performance tests analysing 3D velocities in a point (3D laser Doppler ane- show that for particles located in a 2D plane, the algorithm mometry) or plane (3D stereo-PIV), only a fully 3D tech- can track the particles with a vector yield reaching 100%, nique based on the illumination of a flow volume rather which means that a velocity vector can be determined for than a flow sheet will give the information needed to almost all particles detected. The calculated velocity vec- construct the instantaneous 3D velocity fields. Presently, a tors for this situation have a standard deviation of less few different methods can be applied (Hinsch 1995). The than 1%. The performance is also tested on a mixed first method is an extension of the 3D stereo-PIV tech- convection flow behind a heated cylinder in which the 2D nique. In this method, velocity information in a volume is flow transits into a 3D flow. As there is no exact solution of obtained by acquiring several slices of the flow field using such a flow available, the 3D PTV results are compared a scanning technique (Rockwell et al. 1993; Brücker 1995). with visualisation results. The results show that the 3D The second method is holographic PIV (Zhang et al. 1997), PTV method can capture the main features of the 3D in which the spatial distribution of particles is recorded transition of the 2D vortex street. holographically. The holograms are analysed afterwards by sampling the hologram in slices. A very high spatial res- 1 olution can be obtained in this way, but the experimental Introduction requirements are stringent and the analysis time con- Using tracer particles for fluid velocity measurements is a suming. The third method is a 3D extension of the parti- well-established technique. Tracer particles are seeded in a cle-tracking method. Instead of tracking particles in a thin fluid and illuminated within a defined area. The images of light sheet, the particles are now tracked in an illuminated the moving particles in this area can be recorded and volume. The obtained 3D particle trajectories can be used processed. Currently, several techniques based on particle to calculate the 3D velocity field. Because the actual path of visualisation have been developed to measure the velocity. the particles is analysed, particle-tracking techniques are When the defined area is a thin light sheet, techniques generally more accurate than PIV-based techniques such as 2D particle-tracking velocimetry (2D PTV) and (Cowen and Monismith 1997). Besides, the trajectories of particle image velocimetry (2D PIV) can be applied. In the particles are analysed in a flow volume, which enables, in contrast to scanning techniques, the construction of instantaneous velocity fields. The technique of 3D particle tracking was introduced by Chang and Taterson (1983) Received: 15 September 2001 / Accepted: 13 June 2002 and further developed by (among others) Racca and Published online: 31 July 2002 Dewey (1988) and Maas et al. (1993). In most of these Springer-Verlag 2002 investigations, a 3D particle localisation algorithm is used, based on a so-called epipolar-lines method. In this meth- R.N. Kieft, K.R.A.M. Schreel, G.A.J. van der Plas, C.C.M. Rindt (&) Energy Technology Division, od, several transformations between the physical 3D do- Department of Mechanical Engineering, main and the camera images are needed, resulting in long Eindhoven University of Technology, computational times (for details, see Maas 1996). P.O. Box 513, 5600 MB Eindhoven, The Netherlands In the present investigation, the performance of a 3D E-mail: C.C.M.Rindt@wtb.tue.nl particle localisation and tracking technique is studied and This work is part of the research programme of the Netherlands tested with respect to its accuracy. Both for the calibration Foundation for Fundamental Research on Matters (FOM), which is financially supported by the Netherlands Organisation for Scientific and the accuracy tests, synthesised data is created by Research (NWO). The authors would like to take the opportunity to traversing a precisely manufactured 2D grid through the thank the technical staff of the Energy Technology section for their measuring volume. For the performance test, the fluid flow support. behind a heated cylinder is measured.
In Sect. 2 a thorough discussion of the algorithm is PTV algorithm (as described in van der Plas et al. 1999 and presented showing the details of the 3D localisation pro- Bastiaans et al. 2001). First, the captured images are cedure, the calibration method and the matching algo- dynamically thresholded (Dynamic Thresholding). Then, rithm. The testing problem is discussed in Sect. 3, while within each image, the 2D representation of a particle is the results concerning accuracy and the performance of detected (Blob detection). From the particles located in the the matching algorithm are discussed in Sect. 4. The re- three cameras, a 3D position can be deduced (Mapping to sults of the fluid flow measurements are presented in the Lines of Possible Position and 3D localisation). As soon as Sect. 5. The article ends with a brief discussion and some the 3D position of the particles is known, the procedures conclusions. for matching and path storage are almost equivalent to the ones used in a 2D algorithm. In the following sections, the 604 2 3D localisation of the particles and the 3D calibration Methodology procedure, which are quite different from their 2D repre- In 3D PTV methods, at least two synchronized cameras sentatives, are discussed extensively. The other parts of the need to be used. Only then can stereo images required to algorithm are discussed briefly in appendix A. determine the 3D position of the particles (comparable to the human eye) be obtained. With two cameras, the pos- 2.1 sibility exists that, especially for high seeding densities, Localisation particles are hiding behind each other. As it is the aim to The main idea behind the 3D localisation of particles is track particles as long as possible, the effect of particle that under normal circumstances a particle in a camera ‘hide and seek’ needs to be minimized. Therefore, a third image will have a position in the world coordinate system synchronized camera is applied which focuses at the same somewhere along the particle projection line (Fig. 2a). The volume as the other two cameras (Fig. 1a). A third camera ‘normal-circumstances’ criterion means basically that no also reduces the ambiguity occurring during the 3D abrupt change in refractive index of the medium is allowed localisation. (no ‘mirror’ effects) and that possible changes are known From the obtained images, the 3D position of the par- and stable or predictable in time. This is not a serious ticles can be determined using a 3D localisation algorithm restriction since the applicability of seeding techniques for (Fig. 1b). The algorithm functions comparable to the 2D flow visualisation in general satisfies this criterion. Note that this also holds for different refractive indices (e.g. a Fig. 1a, b. 3D PTV setup and algorithm. a Basic 3D PTV configu- Fig. 2a, b. Principles of the 3D localisation. a Particle projection. ration. b Algorithm scheme for 3D PTV b Method of crossing lines
glass–water interface), or for deforming media (e.g. lens- dependent on the calibration data, the data also deter- like deformations in the glass wall), without affecting the mines whether 3D particles can be found or not. above statement. In the case of a medium with a uniform To obtain the calibration data, a calibration system is refractive index, the lines of possible positions will be developed, allowing to position a precisely manufactured straight. This property is used for the construction of a 2D grid on well-known positions inside the measuring transformation between pixel coordinates of the particle volume (Fig. 3). The calibration grid consists out of a images and the particle projection line in world coordi- blackened square copper-bronze foil (200·200·0.1 mm) nates. When this transformation is known for all cameras, glued onto a flattened opaque glass plate. In this foil, pin- the detected blobs in the images can be transformed to a holes with a diameter of 0.1 mm are etched with a spacing set of lines of possible particle positions (Fig. 2b). The of 5 mm. The grid can be positioned with an accuracy of points in space where the lines from the different cameras ±5 lm. By illuminating the grid from behind with a diffuse 605 cross are the possible positions of the particles. This 3D light source, the regular pattern of pin-holes is visualised localisation method, based on crossing projection lines, and captured by all three cameras. For each grid position, only requires the construction of these lines (Yamamoto the (x,y,z)-positions of the detected pin-holes are exactly et al. 1993). In the so-called epipolar-line method, as used known. This information is used to determine which by various researchers, the virtual images of the particle (x,y,z)-position is projected on each camera pixel. Use is projection lines belonging to cameras 1 and 2 need to be made of nz different z-positions, meaning that each camera constructed in camera 3. This demands an additional pixel corresponds to nz (x,y,z)-positions. By applying a back-transformation from world coordinates to camera least square fit through these points, the coefficients of the coordinates, causing an increase of the computational straight-unique line are established for every cluster of effort. 25 camera pixels. Because blob detection is performed at The ‘crossing-line method’ allows using only two sub-pixel accuracy, a continuous function, describing the cameras. However, in high-seeding-density flows, acci- line coefficients within a camera image, is needed. dental crossing of two lines is quite common, resulting in Therefore, a 2D nf -th-order polynomial (with nf,c-th cross- multiple crossings and consequently in ambiguities in order terms) is fitted through the coefficients belonging to determining the 3D particle position. Besides, particles can each pixel. For this particular experiment, it turns out that also hide behind one another, resulting in no crossings at for nf ‡3 and nf,c ‡3, the error in z-position becomes all. The use of three cameras is therefore almost indis- smaller than the accuracy in the traversing system. pensable. The chances of accidental crossings of three lines through one point are relatively small. 2.3 A blob, detected in each camera image, is now repre- Experimental device sented by three spatial lines crossing somewhere in the The setup consists of a system in which three synchro- measuring volume. Due to optical disturbances, random nized cameras are integrated (Fig. 1a). These three b/w camera noise and errors in the calibration data, an exact CCIR video cameras (JAI 1021, Copenhagen, Denmark) are crossing of the three lines is unlikely. Therefore, a line connected to a RGB frame grabber (PCI frame grabber crossing is detected when the minimum distance between with AM-CLR module, ITI, Indianapolis, Ind.). Each the three lines is smaller than a certain critical value Dc. camera is connected to a different color input of the frame Depending on the quality of the calibration data and grabber and the digitised images are stored as RGB files. camera/lens characteristics, Dc should be set to an optimal Synchronization of the cameras is obtained by using the value. In general, too large a value results in an increase of sink signal from one of the cameras as triggering signal for crossing possibilities, especially for experiments where a the other ones. At the processing stage, the separate high seeding density is used. A very small value, in turn, camera images are then extracted again from the stored results in no crossings at all. In the present study, this RGB images. In this way, a cost-effective setup is realized approaching distance was set to 0.01 mm. with relatively easy synchronization and storage. The frame grabber is installed in a Pentium-class personal 2.2 Calibration For the transformation of the detected blobs in the camera images to lines in world coordinates, in principle a mathematical model can be developed based on the mea- sured geometry of the experimental setup and camera characteristics. The derivation of the transformation pa- rameters in this model, however, would be a time-con- suming task. Furthermore, some kind of check of the transformation has to be performed with in-situ objects. In the present study, in-situ calibration is the preferred technique to determine the transformation. The transformation parameters are determined by tra- versing a well-defined object through the measuring vol- ume. The calibration process should be performed very accurately. Not only the accuracy of the 3D position is fully Fig. 3. Calibration configuration
computer with a 40-GB RAID0 volume (also called stripe system which is connected to the construction which set). The RAID0 volume is created by using the stripe set carries the cylinder. features of Windows NT on ten 4-Gb hard disks, con- By towing the cylinder through the water tank, the flow nected in groups of five to two UW-SCSI controllers. This behind the cylinder is created. When fixing the camera to allows for a minimal sustained data rate of 30 Mb/s, which the setup instead of the translation system, the shed is enough for real-time hard-disk recording with these structures remain within the interrogation volume. cameras. The recording software is Eye-Image calculator Therefore, these structures can be followed for a long (IO Industries, London, Ontario, Canada). Since the period of time. This allows us to monitor the transition cameras are interlaced, the lighting source cannot be process of the shed structures into unstable structures pulsed. Therefore, during the application experiments we where heat effects dominate (Kieft et al. 1999). used a cw Ar+ laser (Spectra Physics 2016, Mountain View, 606 Calif., 6 W all lines) which provides enough intensity to 4 illuminate a typical volume of 10·10·10 cm3. For seeding, Accuracy and performance 50-lm hollow-glass particles are used, which are dispersed The development of the present 3D PTV code can be in the flow about 1.5 h before the actual experiment. thought of as in two parts, the particle-tracking routines and the 3D localisation routines. The integral accuracy of 3 the present 3D PTV method can then be seen as a com- Experimental test problem bination of the 3D localisation accuracy together with the The application on which the 3D PTV code is tested particle-tracking accuracy. In the present article, the concerns a mixed convection flow behind a heated cylin- analysis of the accuracy and performance is focused on the der (Fig. 4). In this problem, a heated cylinder is cooled by 3D localisation. The particle-tracking routine is more or a convection flow. The determining dimensional parame- less independent of whether it is used in two or three ters are then the Reynolds number, ReD=U0D/m, and the dimensions. For more details about the accuracy of the Richardson number, RiD=GrD/Re2D=Dgb(T1–T0)/(U20), tracking routine, the reader is referred to Bastiaans et al. with b the thermal expansion coefficient and m the kine- (2001) in which the accuracy of this part of the method is matic viscosity and the other parameters as defined in discussed. Fig. 4. For small heat addition (small RiD), a 2D von Kärmän vortex street is found, whereas for moderate heat addition, the 2D stable vortex street becomes disturbed 4.1 and a transition to a 3D flow field takes place (Kieft et al. Accuracy in 3D localisation 1999). The performance of the method is first tested for its ca- The experiments were done in a water tank facility. In pability to localise well-known positioned markers some- this setup, the heated cylinder (D=8.5 mm, L=495 mm) is where in the calibrated measuring domain. To that end, towed through the tank rather than being exposed to a the calibration grid is placed at several known z-positions forced main flow. The specific dimensions of the water in this volume. The positions of the pin-holes are now tank are length·width·height=500·50·75 cm. The main reconstructed and compared with the known positions. A advantage of this device is a minimal creation of boundary measure of the average error i in the i-th component of layers and an almost uniform inflow velocity distribution the located pin-hole position vector is the standard devi- (Anagnostopoulos and Gerrard 1978). To obtain the de- ation in the located position with respect to the known sired cylinder wall temperature, an electric rod heater is position. This can be defined according to used with a maximum heat density of 8.0 W/cm2. The sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pnd 2ffi temperature of the cylinder is kept constant in time by k¼1 x i;k ð known Þ x i;k ð located Þ controlling the heat input with the help of the measured ei ¼ nd 1 ð1Þ wall temperature. The translation of the construction is obtained by an electric motor which is corrected for its with xi,k(known) the i-th vector component of the known variation in rotational speed by means of a closed circuit, position vector of the k-th pin-hole, xi,k(located) the i-th resulting in a variation of the rotational speed of less than component of position vector as calculated by the 3D lo- 0.2%. The motor is coupled to the drive wheel by using a calisation algorithm and nd the total number of located 1:100 gear. An almost inelastic fibre-based tape is looped pin-holes. around the drive and the idle wheel of the translation It turns out that the error in the located positions i is dependent on the camera configuration (Fig. 5a). The camera configuration is determined by the camera viewing direction and the optical path lw through the water. The camera viewing directions a1,2,3 are defined as the angles between the optical axis of the cameras and the z-direction (central line in Fig. 5a) after refraction by the air/test- section interface. The dependency of i on these variables is investigated by three sets of experiments, where in the first set, the camera viewing angles are a1=a2=a3=a. In the second set, Fig. 4. Definition of the mixed-convection problem a1a25 and a3 is varied, while in the third set, lw is
deformation of the images at the air/water interface when viewing with an angle. For the second set (denoted by a triangle and scaled with the upper axis, Fig. 5b), in which only a3 was varied, an increasing a3 also results in a decrease of z, but less strongly than for the first set. Furthermore, at a3=25, the accuracy of the z-position is still improving, while for a varying a, this improvement seems to diminish for a>20. The influence of the optical path length is determined for a13 and by varying the position of the measuring volume within the test section (denoted by a square and 607 scaled with the middle axis, Fig. 5b). By doing so, it turns out that this variation has no detectable effect on the ac- curacy in z-position. Only a slight improvement can be observed if the measuring volume is positioned closer to the air/water interface. Considering the camera positions, the ideal configura- tion is an orthogonal setting, resulting in a set of images with the smallest dependency. From the results one can conclude that the smallest error is observed for viewing angles larger than 20 with regard to the central line. No significant improvements in accuracy can be gained by a further increase of this angle. The remaining error in z-position z is approximately 0.003 cm. Comparing this accuracy with the accuracy in x- and y-position, it is found out that z is about three to four times larger than x,y. This accuracy is similar to values found by other researchers (Sata and Kasagi 1992). 4.2 Tracking performance The 3D tracking of particles is tested for the optimal camera configuration (a20, lw150 mm). In this test, the calibration grid is translated through the measurement Fig. 5a, b. Performance variables and their influence on the domain with a constant step size Dz=0.1 cm. As in this test z-position. a Camera configuration. b Dependency of the accuracy in all particles are situated in a plane, no ambiguities during z-position on the camera configuration. Three tests were performed the localisation (occurrence of two crossing possibilities) (see text for definitions) or the phenomenon of ‘hiding’ particles occur. Further- more, only about 200 white dots are viewed by the cam- eras, corresponding to a low seeding density. At every varied. During these experiments, the accuracy of the in- position of the grid, three independent camera images are stalled angle was estimated to be 1 and the accuracy of the captured, from which the pin-hole positions are deter- measured optical length was 0.5 mm. mined. Doing this for a sequence of ten steps, a tracking For all three sets, it was found that x and y are more or run is simulated where a virtual velocity can be detected less independent of the varied variable and equal to about between two subsequent frames. The accuracy of the 10 lm. The error in z-position z, in turn, strongly varied matching in z-direction can then be determined by plot- as a function of the chosen parameters (Fig. 5b). For a ting the distribution of all found z-velocities during the changing a (depicted as the black circles and scaled ac- entire tracking run (Fig. 6). These z-velocities show only a cording to the bottom axis), the strongest variation in z small standard deviation of about 1.2% of the mean value. can be observed. For increasing a, z decreases rapidly for In a real flow, the total number of localised particles is 8
position behind the cylinder, 3D phenomena will arise. The 2D vortex tubes shed from the cylinder will then show a 3D transition, eventually leading to a collapse of the vortex street. This transition is analysed here by consid- ering the z-component of the vorticity vector xz (Fig. 7- a,c,e). The evolution of xz is interpreted using the visualisation results. The results show indeed that the flow remains 2D over a certain period after the vortex structures are shed. The isovorticity surfaces (here xz=–0.3 and xz=0.3, Fig. 7a) appear as more or less axisymmetric vortex tubes, which 608 are parallel to the cylinder axis. The 2D character of the flow can also be seen in the dye visualisation results (Fig. 7b). A regular pattern of dye streaks is observed up to x/D
609 Fig. 7a–f. Isovorticity surfaces [xz=–0.3 (shaded) and xz=+0.3 (grid)] as determined by 3D PTV (left) and visualisation results (right) for ReD=75 and RiD=1.3 measuring volume is at least 15. For viewing angles above • In the 3D localisation algorithm used, only the first 25, no significant improvement in localisation accuracy found crossing possibility is used. Other crossing was found. possibilities are not taken into account and this From the 3D PTV results it is concluded that the information is lost. During the particle matching, an method is capable of detecting a thermally induced tran- erroneous crossing may be chosen. This will result in sition of a stable vortex street. In these experiments, about no matching with previously located particles and 500 vectors were constructed, allowing the calculation of therefore cause a decrease in the velocity yield. At this the vorticity distribution. The 3D PTV results closely re- point, the algorithm can be improved by using all semble the results observed by visualisation experiments. crossing possibilities for matching. The incorrect Although not presented in this article, the results also crossings then do not match with other particles and showed that, for an increased seeding density, the number automatically drop out during the matching of resulting vectors drops significantly with respect to the procedure. found blobs. In comparison to other reported results • For a high seeding density, the possibility exists that two (Maas 1996), this decrease is quite common and is mainly or more particles are located on one particle projection caused by the strong increase of ambiguities for increasing line (hiding particles, Fig. 2b). This results in only one seeding density. In order to obtain a higher vector yield, particle image in the camera frame. Within the present the present algorithm can be improved at least with algorithm, every 2D localised blob can only be used respect to the following points. once, resulting in a loss of 3D particles. Allowing mul-
tiple usage of a 2D blob will overcome this problem and After this validation step, the particle positions are consequently result in an increased number of 3D determined with sub-pixel accuracy by using the localised particles. grey-value-weighted centre of gravity (volume centroid) of the segmented blob. The above-suggested improvements are currently being implemented in the 3D PTV code. 3D localisation The localization method is thoroughly described in the main text of the article. Appendix: Description of the interrogation procedure Matching Dynamic thresholding 610 Every particle in frame f is matched with a candidate In order to accurately determine the position of a par- particle in frame f+1, where the candidate particles are ticle, the background light (due to the reflections and the defined as all particles within a given maximum matching non-uniform intensity profile of a light sheet) is re- distance from the particle in frame f. From all candidate moved. A simple and fast algorithm was chosen: a particles, the one which is positioned closest to the po- square min-max subtraction filter. The filter leaves the sition of the particle in frame f+1 is matched (Matching). intensity profile undisturbed while removing non-uni- The matching algorithm is improved by using an esti- formities of its filter size. The filter consists of three mation for the particle position in frame f. This estimated basic operations: position is provided by a prediction algorithm (Predic- • Step 1: Min. Each pixel in a copy from the original image tion). In the 3D PTV algorithm, the prediction is deduced is replaced by the minimum value in a square filter from the flow calculated in the frame set f–1. An im- window. portant parameter for matching is the maximum • Step 2: Max. Each pixel in the Min-filtered image is then matching distance. As the cameras operate at 25 Hz and replaced by the maximum value in a square filter win- the average flow velocity is about 1 cm/s, the maximum dow. matching distance was set to 0.2 cm (equivalent to • Step 3: Sub. The result of step 2 is then subtracted from 10 pixels). the original image. References A rectangular window was chosen for the filtering. This Agui C, Jimenez J (1987) On the performance of particle tracking. filtering by rectangular windows is obtained by a line filter J Fluid Mech 185:447–468 in the x-direction followed by a line filter in the y-direc- Anagnostopoulos E, Gerrard J (1978) A towing tank with minimal tion. The window width and height should correspond background motion. J Phys E 9:951–954 Bastiaans R, Plas G van der, Kieft R (2001) The performance of a new with the maximum pixel width and pixel height of particles PTV algorithm applied in super-resolution PIV. Exp Fluids one wishes to detect. During the filtering procedure, a filter 31:346–356 size of 20·20 pixels was used. Brücker C (1995) Digital particle image velocimetry (DPIV) in a scanning light-sheet: 3D starting flow around a short cylinder. Exp Fluids 19:255–263 Chang T, Taterson G (1983) Application of image processing to the Blob detection analysis of three-dimensional flow fields. Opt Eng 23:283–287 The algorithm for blob detection is quite simple. Pixels Cowen E, Monismith S (1997) A hybrid digital particle-tracking ve- that have a vertical or horizontal neighbour are called locimetry technique. Exp Fluids 22:199–211 connected to each other. Regions of connected pixels in Dalziel S (1993) Decay of rotating turbulence: some particle-tracking experiments. In: Nieuwstadt F (ed) Flow visualization and image an image with an intensity higher than a threshold level analysis. Kluwer, Dordrecht, The Netherlands, pp 27–54 are called blobs. The single threshold for blob detection Hinsch KD (1995) Three-dimensional particle velocimetry. Meas Sci is set slightly above the noise level. Since the images are Technol 6:742–753 dynamically thresholded to remove background varia- Kieft R, Rindt C, Steenhoven A van (1999) The wake behaviour behind a heated horizontal cylinder. Exp Therm Fluid Sci tions, a single threshold for blob detection is sufficient. 19:183–193 The threshold level was set to a level of 20 grey values. Maas H (1996) Contributions of digital photogrammetry to 3d ptv. In: The detected blobs fulfilling certain shape and intensity Dracos T (ed) Three-dimensional velocity and vorticity measuring criteria are then called ‘blobticles’ (valid particle and image analysis techniques. Kluwer, Dordrecht, The Nether- images). lands, pp 191–207 Maas H, Gruen A, Papantoniou D (1993) Particle-tracking velocimetry Only minimum required blob size and maximum al- in three-dimensional flows. Exp Fluids 15:133–146 lowed blob size were implemented. Small particles as used Malik N, Dracos T, Papantoniou D (1993) Particle-tracking velocim- for tracking will appear round with a Gaussian-like in- etry in three-dimensional flows. Part II: Particle tracking. Exp tensity profile due to the characteristics of the imaging Fluids 15:279–294 Nichino N, Kasagi N, Hirata M (1989) Three-dimensional particle- optics. In special cases where noise, reflections, and small tracking velocimetry based on automated digital image process- background effects remain after dynamic thresholding, ing. ASME J Fluids Eng 111:384–391 one can easily add shape and intensity criteria as needed. Plas G van der, Kieft R, Rindt C (1999) Application of a 2D high- In Dalziel (1993), for example, criteria like maximum el- resolution particle veclocimetry method on mixed convection flows. In: Adrian R, Hassan Y, Meinhart C (eds) Third Int. lipticity and minimum required averaged intensity are Workshop on PIV, 1999, pp 177–182 used to distinguish real particle images from particle-sized Racca R, Dewey J (1988) A method for automatic particle tracking in a intensity blobs. three-dimensional flow field. Exp Fluids 6:25–32
Rockwell D, Magness C, Towfighi J, Akin O, Corcoran T (1993) High- Yamamoto F, Uemura T, Iguchi M, Ohta J, Wada A, Mori K (1993) 3D image-density particle image velocimetry using laser scanning PTV based on binary image correlations method and its applications techniques. Exp Fluids 14:181–192 to a mixing flow with a bubbling jet. In: Brebbia CA, Carlomagno GM Sata Y, Kasagi N (1992) Improvement toward high measurement (eds) Computational methods and experimental measurements VI. resolution in three-dimensional particle tracking velocimetry. In: Springer, Berlin Heidelberg New York, pp 229–246 Tanida Y, Miyashiro H (eds) Flow visualization VI. Springer, Zhang J, Tao B, Katz J (1997) Turbulent flow measurement in a square Berlin Heidelberg New York, pp 792–796 duct with hybrid holographic PIV. Exp Fluids 23:373–381 611
You can also read