Open the 'civ' interface and enter the .avi file as input, even if you cannot read them on your computer: the programme will need it to get the number of frames and time interval (matlab command 'aviinfo'). Don't open the .png images as input: it would lead to errors since they don't form a continuous sequence.
First pick up a reference image in the series, for instance number 1001, so we display 'image #' 1001 ('first') 1 ('increment') 1001 ('last'). Many image pairs can be chosen around this reference image, as displayed in the menu 'CIV1_PAIR'. With the choice '0|1', the reference image (0) is compared to the next one (1). '-1|1' compares the previous image to the next one, so the time interval is double, and so on. The corresponding time intervals are indicated in ms. As a first try, let us choose a small time interval, 160 ms, corresponding to '-2|2'. This insures a good correlation between the images (in this example, the exposure time covers four images, to improve the luminosity, so images change only every four frames and we need to choose a time interval multiple of four).
The resulting velocity field 'UV_DEMO/EX_2/A/movie_999-1003.nc', can be displayed by 'uvmat'. The velocity field looks good, with blue vectors indicating excellent image correlation, but the corresponding displacement in pixels, as seen in the histograms, only ranges from -2 to 1 for the u component and even less for the v component. Since the typical precision on displacement measurement is 0.2 pixels, this means that the relative precision is low, of the order of 10%.
Let us now make the opposit test, running civ again with a time interval ten times longer, with the pair -20|20. We now observe many green arrows, indicating that the image correlation is less good, and some vectors are erratic. We also observe 'black' vectors, corresponding to warning flags. The flag can be identified by selecting the vector with the mouse, after pressing 'select' on the upper right. Then the vector index and coordinates appear in the upper text display window, followed by F=-2. This indicates that the search range is too small: the correlation maximum is found at the edge of the search range. We indeed notice that the black vectors correspond to large velocities. A complete description of flags can be found in the online help (press the online 'HELP' near the 'ACTION' ). It is possible to fix this problem by increasing the search range in the civ interface ('isx,isy'). However it is more reliable to use a smaller time interval, for instance the pair -8|8, and possibly increase it with the civ2 processing, as described below.
|
[time interval too short (pair-2|2). The correlation is good, as indicated by the 'blue' vectors, but the pixel displacement shown by the histogram, is limited to +-2 pixels.]
[Good choice (pair -8|8). The histogram is fairly wide and is not constrained at the bounding values.]
[Time interval too long (pair -20|20). The correlation is not very good, as indicated by 'green' vectors, and the search range is too short, as indicated by the 'black' vectors and the histogram constrained at the lower bounding value.]
|