DigImage: Particle Tracking
:::: IMAGE CAPTURE
:::: PARTICLE LOCATION
:::: PARTICLE MATCHING
:::: STRUCTURE OF FILES
:::: BASIC OPERATION OF TRK2DVEL
:::: CONFIGURATION OF TRK2DVEL
:::: PARTICLE-TAGGED DATA
:::: IMAGE PREPROCESSING
Prior to tracking an experiment it is necessary to set up a range of parameters which control the tracking process. A complete description is beyond the scope of this document, but they may be classified into six main groups: coordinate system specification, video control, image enhancement, particle location, particle association and data output. Some of these parameters have already been discussed in this document. More complete details may be obtained from the help text within the various submenus involved.
As DigImage utilises the world coordinate system, it is essential that this be defined before the tracking proceeds. This is achieved interactively using the facilities in [;PWW Determine world coordinate mapping] (the list should first be initialised by [;PWV Initialise world coordinate mapping]) by locating a number of features on an image (typically a grid temporarily placed in the experimental apparatus) and specifying their world coordinates. A variety of mapping functions are then available to fit a mapping between the world and pixel coordinate systems. The simplest function consistent with the optical arrangement should generally be chosen. Typically this will involve a simple linear mapping between pixel and world coordinate spaces.
A second stage of mapping is desirable as there are generally small discrepancies in the image location between one frame and the next due to small synchronisation and timing errors in the video signal. When provided with a set of permanent reference points, DigImage is capable of automatically generating this mapping for each sample it processes. These reference points will typically consist of small points of light on a dark background (or dark dots on a light background) arranged in a vertical strip on either side of the viewed area, just outside the window to be tracked. It is preferable to avoid having particles visible in the immediate vicinity of the points; if they are visible, the points must be clearly distinguishable by a threshold higher than the maximum expected for any particle. Each point must be registered with DigImage (using [;PRL Locate permanent reference point]), along with the threshold defining it, and the mapping function selected (by [;PRM Mapping type for perm cal points]). At the start of the tracking process, the points previously registered with DigImage will be located automatically, and their current locations used as the reference position (instead of their original registered positions).
For subsequent samples, the new position of the points will be found and a mapping evaluated to transfer any data from the coordinate system in the new frame to the coordinate system in the initial frame before the data is converted to world coordinates. If the root mean square (rms) error in the mapping function exceeds some specified value, then DigImage will force the image to be captured again from video in an attempt to improve the accuracy of the mapping. If after five retries an acceptable rms error value is still not achieved, then DigImage will flag an error, but continue with the tracking process.
Control of the video hardware includes specification of the maximum time over which particles are to be tracked, an the sampling frequency. The tracking may be either forwards or backwards in time (for a flow in which the flows are nearly singular at some point, tracking backwards towards that point may be more effective than tracking forwards from it), with sampling at up to 50Hz (60Hz for NTSC). Note that when sampling at the maximum frequency there will be a small error in the vertical positioning of particles due to the interlacing of the video lines. The vertical resolution will also be reduced from 512 to 256 lines (480 to 240 lines for NTSC). Sampling at this maximum frequency only makes sense when a field integration mode is used. Refer to Document\Cameras.DOC for details.
Some of the techniques available for image enhancement by removal of variations in the background illumination have been discussed already. In addition to these a range of filter options are available to clean up the image in a variety of ways, such as low pass filtering. When dealing with high velocities (in pixels per sample), the 1/50 (1/60 for NTSC) time difference between the information contained on the even and odd lines, due to the interlacing of these lines, can lead a particle to appear to be in two places at once. In such circumstances it is necessary to discard half the information present in the image and utilise only the even or the odd lines (not both, unless tracking at the maximum frequency which corresponds to single video field spacing of the samples), and operate with a vertical resolution of 256 pixels. Filters exist for a variety of methods of allowing for the interlace, depending on the velocities concerned (see [;USI: Filter type] for more details). Note that if you have a frame integration camera in combination with a mechanical shutter, it is possible to avoid these problems and at the same time achieve a significantly improved resolution. For further details refer to Document\Cameras.DOC.
Parameters controlling particle location (in [;USL: Particle location]) specify the two limiting thresholds and number of thresholds used in their location, limits on the size and ellipticity of the particle (beyond which it may be treated as two particles), the centroiding technique to be used to define the particle location, and the maximum allowable difference between the two centroids. Some of these location parameters interact with the costing parameters (in [;USP: Pricing policy]) which control the association process. The three most fundamental costing parameters specify the form of the costing function to be used ([;USPP: Policy type]), the maximum "error" between the predicted position of a particle and its actual position before the two images will be treated as separate physical particles ([;USPM Maximum matching distance]), and the maximum velocity error a new particle entering the tracking region may have ([;USM Max velocity error for new paths]). Some of these parameters are described in more detail in an earlier section. More detailed information may be obtained from the help facility associated with the relevant options.
Results of the tracking process are output both visually
and in machine readable form. The visual output provides a number
of ways of viewing the particle streaks in addition to diagnostics
on the tracking status and cost matrix (controlled by [;USD
Display particle paths]).
Four permanent files are created by DigImage. The first (base_name.PAR)
all the tracking parameters used to control the tracking, while
the second (base_name.WLD)
details on the coordinate system used. The third file (base_name.IND)
as an index to the main tracking file. Finally the main tracking
direct access binary file which may be many Mbytes in size. When
tracking the maximum number of particles (at present 4095) at
25Hz, over 36MBytes (each sample requires 80 bytes plus 6 bytes
per particle in the sample) of data is produced for each minute
of experimental time. To operate the particle tracking system
effectively it is therefore necessary to have a very large hard
disc available, especially if ensemble statistics are required.
DigImage documentation page
DigImage home page
Stuart Dalziel's home page