From: Cline, Danelle
Sent: Thursday, April 28, 2005 3:19 PM
To: Edgington, Duane; Oliver, Dorothy; Davis, Dan
Subject: AVED classifier matlab code changes and usage


Hi AVED fans, 

Below is a summary of the recent Matlab changes and instructions on usage of the functions. I’ve completed all changes that I think are necessary for the upcoming poeobius tests. Please let me know if you have any problems or questions.

Summary of changes:

I modified the background subtraction algorithm. It similar as it was before, except it can now handle small images without crashing.  I also added an "event classifier" module that has both a majority win, and probability win methods (see step 3 below for usage). I lastly added support for passing separate arguments for image and database files  to make the file management easier.

To get the latest matlab code, checkout the aved/matlab module in CVS.  

The steps and usage of the matlab code:

1. Build your training class with build_classes

Usage: build_classes <class image directory> <database directory >

Example: build_classes /mnt/scratch/doliver/poebius/PBSbackground ./DB/PBSbackground

This function builds a training set of classes. If the <class image directory> has subdirectories, it reads all the subdirectories and create classes with the subdirectory names. If no subdirectories exist, it creates a class with the parent directory name.  In the above example, this would train a single class called PBSbackground. Also, please note that this function creates the same directory structure (e.g. features/classes, features/tests, etc.) and will store all data in *.mat files just as before.

2. Collect your test data with collect_tests

Usage: collect_tests <test image directory> <database directory>

Example: collect_tests /mnt/scratch/doliver/poebius/dive2613 ./DB/PBSbackground

This function collects the image data to test against the class data. In this example it would collects dive2613 data and save the collected data in ./DB/PBSbackground. This function also creates the same directory structure (e.g. features/classes, features/tests, etc.) and will store all data in *.mat files just as before.

3.  Run your test data against the class data with run_tests

Usage: run_tests <database directory>  <matlab results file>  <(optional) majority|probability>

Example: run_tests ./DB/PBSbackground pbstestbackground majority probability

This function will test the collected test data with the class built in step one. In the above example, this would classify with both majority and probability methods. The resulting test data would be saved to pbstestbackground.mat, pbstestbackground-majority.mat, and pbstestbackground-probability.mat. These three files represent the single frame classified results and the event classified results for both majority and probability rules. 

Please let me know if you have any questions.

======================================
Danelle Cline, Software Engineer, R&D Eng x1947
                                           O  o    
                                         _\_   o
                                      \\/  o\ .
                                      //\___=
                                         ''