[Robotflow-cvs] RobotFlow/demo/SymbolRecog README.TXT,1.2,1.3

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 422-6466

Update of /cvsroot/robotflow/RobotFlow/demo/SymbolRecog
In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv8455

Modified Files:
	README.TXT 
Log Message:
Updated demo

Index: README.TXT
===================================================================
RCS file: /cvsroot/robotflow/RobotFlow/demo/SymbolRecog/README.TXT,v
retrieving revision 1.2
retrieving revision 1.3
diff -C2 -d -r1.2 -r1.3
*** README.TXT	23 Aug 2004 23:36:37 -0000	1.2
--- README.TXT	4 Nov 2004 18:51:33 -0000	1.3
***************
*** 2,27 ****
  ----------------

! This demo contains FlowDesigner networks that are used to 
! recognize printed text (alphanumeric) extracted from live (real-time) 
! 320x240 color images. FlowDesigner networks provided with this demo are 
! easy to modify to suit your configuration (image size, video capture hardware).
! The idea with this demo is to provide a simple algorithm that extracts characters 
! from color components. Characters are extracted from the image with a simple 
! color segmentation  algorithm (8 neighbors), with 1 color representing the background and 
! 1 color representing the foreground color (printed text). When probable characters are 
! found in the image, the algorithm gives incremental pan-tilt-zoom commands to the 
! camera (if supported by the camera) to get the maximum resolution of the characters 
! to extract. When the size of the characters is adequate, components are scaled and 
! sent to a feed forward artificial neural network (ANN) on character at a time. The output 
! of the ANN, for each character, is used as probability for each character to be part 
! of any word in the dictionary.

- Please read section 7.0 to get started with the demo. 

  2.0 PUBLICATIONS
  ----------------

! Lï¿½ourneau, D., Michaud, F., Valin, J.-M., Proulx, C. (2003), "Textual message read by a mobile robot", 
! Proceedings IEEE/RSJ International Conference on Intelligent Robots and Systems, p. 2724-2729

  Available at : 
--- 2,28 ----
  ----------------

! This demo contains FlowDesigner networks that are used to recognize printed text
! (alphanumeric) extracted from live (real-time) 320x240 color images.
! FlowDesigner networks provided with this demo are easy to modify to suit your
! configuration (image size, video capture hardware).
! The idea with this demo is to provide a simple algorithm that extracts
! characters from color components. Characters are extracted from the image with
! a simple color segmentation algorithm (8 neighbors), with one color representing
! the background and one color representing the foreground color (printed text).
! When probable characters are found in the image, the algorithm gives incremental
! pan-tilt-zoom commands to the camera (if supported by the camera) to get the
! maximum resolution of the characters to extract. When the size of the characters
! is adequate, components are scaled and sent to a feed forward artificial neural
! network (ANN) on character at a time. The output of the ANN, for each character,
! is used as probability for each character to be part of any word in the
! dictionary.

  2.0 PUBLICATIONS
  ----------------

! Letourneau, D., Michaud, F., Valin, J.-M., Proulx, C. (2003), "Textual message
! read by a mobile robot", Proceedings IEEE/RSJ International Conference on
! Intelligent Robots and Systems, p. 2724-2729

  Available at : 
***************
*** 35,39 ****
  4.0 LICENSE
  -----------
! LGPL, please visit http://www.gnu.org/copyleft/lesser.html for the full description.

  5.0 REQUIREMENTS
--- 36,41 ----
  4.0 LICENSE
  -----------
! LGPL, please visit http://www.gnu.org/copyleft/lesser.html for the full
! description.

  5.0 REQUIREMENTS
***************
*** 44,50 ****
  - wget (to download the templates)

! You will also need to get the template images (required once) to get the full demo
! working properly. Simply run the getTemplates.sh script, it will download the required files
! and decompress them to the templates directory 2Mb download ( + 14Mb once decompressed).

  6.0 PROVIDED FILES
--- 46,57 ----
  - wget (to download the templates)

! You will also need to get the template images (required once) to get the full
! demo working properly. Simply run the getTemplates.sh script, it will download
! the required files and decompress them to the templates directory 2Mb download 
! (+ 14Mb once decompressed).
! 
! Note : This demo is intended to work with a Sony SNC-RZ30 network camera.
! However, any V4L2 capture card with a PTZ could be used. The user must then
! modify the demo to suit its hardware. Please contact us if you need assistance.

  6.0 PROVIDED FILES
***************
*** 58,61 ****
--- 65,69 ----
  |	|- testTraining.n (test the neural network training)
  |	|- trainNeuralNet.n (train the neural network)
+ |	|- colorTrain_SNCRZ30.n (train the color lookup)
  |- templates
  |	|- README (Templates README)
***************
*** 73,79 ****
  7.0 HOW TO USE THE DEMO
  -----------------------
! After you have installed all the softwares listed in Section 5, you are ready to start the demo.
! There are 3 steps to the procedure, but the first two are optional, since the trained ANN are provided 
! with this demo. Training the neural networks may take 1-2 hours.
  From the demo/SymbolRecog directory :
  2) Download the templates with the script getTemplate.sh (optional)
--- 81,89 ----
  7.0 HOW TO USE THE DEMO
  -----------------------
! After you have installed all the softwares listed in Section 5, you are ready to
! start the demo.There are 3 steps to the procedure, but the first two are
! optional, since the trained ANN are provided with this demo. Training the neural
! networks may take 1-2 hours.
! 
  From the demo/SymbolRecog directory :
  2) Download the templates with the script getTemplate.sh (optional)
***************
*** 81,93 ****
  4) Run the live demo with flowdesigner :
  	- % flowdesigner SymbolTracking.n
! Comments are inserted into the flowdesigner networks so you can understand what's happening 
! and make modifications to them according to your hardware and your needs. 

  8.0 HOW TO TRAIN NEURAL NETWORKS
  --------------------------------
  1) Go in the n-files directory
! 2) You must classify all the templates before you can train them (you only need to do that once)
  2.1) To classify the templates run : % gflow classifyTemplates.n
! 2.2) You should see a GUI with a letter or digit appearing in the ImageProbe ("J" is the first one).
  2.3) Enter the corresponding letter or digit in the SymbolKeypad
  2.4) Click Forward on the probe "CLICK_FORWARD_FOR_NEXT_IMAGE"
--- 91,109 ----
  4) Run the live demo with flowdesigner :
  	- % flowdesigner SymbolTracking.n
! Comments are inserted into the flowdesigner networks so you can understand
! what's happening and make modifications to them according to your hardware and
! your needs. Symbols (text) must be printed black on orange sheet of paper for
! this demo to work properly. You could also change the color lookup files by
! running : 
! 	- % flowdesigner colorTrain_SNCRZ30.n

  8.0 HOW TO TRAIN NEURAL NETWORKS
  --------------------------------
  1) Go in the n-files directory
! 2) You must classify all the templates before you can train them (you only need
! to do that once)
  2.1) To classify the templates run : % gflow classifyTemplates.n
! 2.2) You should see a GUI with a letter or digit appearing in the ImageProbe
! ("J" is the first one).
  2.3) Enter the corresponding letter or digit in the SymbolKeypad
  2.4) Click Forward on the probe "CLICK_FORWARD_FOR_NEXT_IMAGE"
***************
*** 96,104 ****
  	- ../templates/train_in_172_v2.data
  	- ../templates/train_out_172_v2.data
! 3) The files will contain all data required to train the neural network (input & output vectors)
  3.1) Training the network requires to run the following flowdesigner network :
  	% gflow trainNeuralNet.n
! 3.2) You can also test the neural network by running the following flowdesigner network :
! 	% gflow testTraining.n

--- 112,122 ----
  	- ../templates/train_in_172_v2.data
  	- ../templates/train_out_172_v2.data
! 3) The files will contain all data required to train the neural network (input &
! output training sets (vectors) )
  3.1) Training the network requires to run the following flowdesigner network :
  	% gflow trainNeuralNet.n
! 3.2) You can also test the neural network by running the following flowdesigner
! network :
! 	- % gflow testTraining.n