Characterizing Particle Tracking

Overview of Approach

The general approach taken here is to provide a set of image sequences, description files, and tracking logs that can be used to validate algorithms that track particles over time.

Each image sequence represents a time series of motion of either a synthetic or an actual data set of particles moving in 2D images. Each sequence is available either as an 8-bits-per-color AVI file or as a set of 8-bit-per-color TIFF files or as sets of 16-bit-per-color TIFF files. Each is monochrome (either stored as a single channel or stored with all color channels having equal values at each pixel).

There is an HTML description file associated with each image sequence that describes in detail how the data set was synthesized or collected. These files are intended to be as complete as possible, describing all known noise in the image, what is being viewed, how it is being viewed, and what is causing motion. The description file also tells how the data from each tracking log was collected, including a discussion of error sources accounted for; for each log, an image is included that indicates the particular particle being tracked.

One or more tracking logs are associated with each each image sequence. Each tracking log is an ASCII-format comma-separated-values file that describes the motion of a single particle within the associated sequence. Each has a header that describes what the columns are. Positions in the columns are in units of pixels, with coordinate (0,0) located at the center of the pixel at the lower left-hand corner of the image and (NX-1, NY-1) located at the center of the upper right-hand corner of the image. Each pixel is one unit wide, making the coordinate of the lower-left-hand edge of the image (-0.5, -0.5) and the coordinate of the location halfway between the first and second row and column be (0.5,0.5). Some logs are relative logs, where the initial position of the particle is unknown; in this case, the initial bead position is recorded as (0,0) and the later reports are offsets from this initial position. No absolute bead-tracking logs begin with the particle located at (0,0).

Tables of results

The following results are for Video Spot Tracker version 08.00:

Effect of Precision and Sample Spacing on Accuracy and Speed

Effect of Tracker Radius on Accuracy and Speed

Effect of Contrast and Noise on Accuracy and Speed

Effect of Noise on Accuracy and Speed

Effect of Bead Size on Accuracy and Speed

FIONA kernel test against Airy disc simulation

Effect of Kernel type on Accuracy and Speed

Effect of Noise on Accuracy and Speed in rods

Effect of Contrast and Threshold on Auto-find Performance

Effect of Noise on Auto-find Performance

Effect of Blur Setting on Auto-find Performance

Effect of Center Surround Setting on Auto-find Performance

What is Characterized

These video sequences are designed to characterize error, precision, bias, and speed. All particle trackers estimate the position in X and Y. Some estimate orientation. Some estimate radius. Some estimate position in Z. The results are reported separately for each.

XXX How do we estimate "RMS", and should it be included separately? XXX How do we determine if the algorithm lost tracking -- the sorts of things that make it lose tracking (noise, many nearby particles, too-rapid particle motion, changes in particle characteristics)?

Manual

There is a manual available for running the test code itself available here.