This README written by Steven Beauchemin

This program, nagel.c, is an implementation of Nagel 1987's motion
detection algorithm.
To compile:

   make -f makenagel

Everything is in a single file so a make is really not necessary but...
This program uses a sequence of input images and smooths them according
to the sigma values provided by the user prior to the estimation of the optical 
flow using a 2nd-order derivative technique. 1st order intensity derivatives are 
computed using 4-point central difference masks and 1st order velocity derivatives
are computed using 2-point central differences. 2nd order derivatives are computed
as cascaded of 1st order derivatives. An oriented smoothness relaxation scheme
is also applied (see Nagel's 1987 paper)

Examples of run files are also present. (You'll have to change the directory 
names where the input and output are directed.) Here's a description of a 
typical command line (that is exactly what one gets if one invokes the 
program with no parameters):

Usage: nagel input-path output-path seq-name. [-SG n.n] [-TG n.n] [-M n] [-I n] [-A n.n] [-F n.n] [-C corr-vel-file histo-data-file -T nbins increment]
[-SG n.n]        : sigma value for spatial gaussian
                   default value is 3.0
[-TG n.n]        : sigma value for temporal gaussian
                   default value is 1.5
[-M n]           : middle frame of image sequence
                   default value is 0
[-I n]           : number of relaxation iterations
                   default value is 10
[-A n.n]         : smoothing parameter alpha
                   default value is 5.0
[-F n.n]         : filters out unreliable estimates
                   using the magnitude of local gradient
[-C corr-vel-file -T nbins increment]
                 : error histogram
                   corr-vel-file : file of correct velocities
                   nbins : number of bins for histogram
                   increment : value of gap between bins

Here is a typical run:
 
nagel /data/images1/BURKITT/testdata /data/images1/BURKITT/outdata newbinarytreet. -SG 1.5 -TG 1.5 -M 20 -A 0.5 -I 100 -B 150 150 -F 5.0 -C correct_trans20 20 0.5 

This command means: run uras using the input files from 
/data/images1/BURKITT/testdata and put output files in
/data/images1/BURKITT/outdata 

newbinarytreet. is the stem name of a typical set of input images (for example,
the complete filename for frame 9 is "newbinarytreet.9"). In the above example,
the program expects to find the input images in /data/images1/BURKITT/testdata.
The data is in binary format (1 character per grayvalue) so the -B option tells
the program that the images are 150 by 150.

-SG 1.5 indicates that the sigma value of the spatial smoothing of the image
sequence is 1.5. Typically this has to be higher for real image data, say in the
range of 3.0 to 5.0.

-TG 1.5 indicates that the sigma value of the temporal smoothing of the image
sequence is 1.5 (the program computes the required number of frames by itself,
see option -M)

-M 20 indicates that the flow field will be estimated for frame 20, the
program will automatically read in the required number of frames on both 
sides of frame 20, according to the sigma value of temporal smoothing given
by -TG and the size of the derivative masks.

-A  0.5indicates that value 0.5 is used for ALPHA in the relaxation 
process. Typical values range from 0.5 to even 100.0. (a -D option also
exists which allows one to set the DELTA paramter. This has a default
value of 1.0.)

-I 100 indicates that 100 relaxation iterations will be executed

-C indicates the correct answer is available and that you want error analysis
conducted and that data is in /data/images1/BURKITT/testdata/correct_treet 
(change this as well as appropriate). The data is analysed by producing a
histogram. The next two numbers specify the number of bins in
the histogram and the increment between them. The histogram is
constructed by considering the error of the estimates versus the magnitude
of the local gradient

-F n.n allows the user to preserve the estimates that show 
local gradient >= n.n only. (If this option is not present no thresholding
is performed.)

The program automatically generates the output filenames for the flow field
and the histogram (if an histogram is requested). For the above example,
the filename of the flow field would be:

nagel.newbinarytreet.F-sg1.5-tg1.5-m20-a0.5-i100-f5.0

and the filename of the histogram:

nagel.newbinarytreet.H-sg1.5-tg1.5-m20-a0.5-i100-f5.0

Note that the letters and numbers of the extension are the parameters 
used when invoking the program.

The program takes between 0:30 and 15:00 minutes to run on an IBM 6000,
depending on the size of the images being processed and the nuber of
relaxation iterations to be performed. See the README for
information about the output format for flows and use software in ../SOFTWARE 
to print out the flow fields in postscript.