+++ /dev/null
-<HTML>\r
- <HEAD>\r
- <META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">\r
- <META NAME="Generator" CONTENT="Microsoft Word 97">\r
- <META NAME="Template" CONTENT="D:\WIN32APP\msoffice97\Office\html.dot">\r
- <META NAME="GENERATOR" CONTENT="Microsoft FrontPage 4.0">\r
- <META NAME="Author" CONTENT="Gary Bradski">\r
- <META NAME="Description" CONTENT="Header for main functions">\r
- <META NAME="KeyWords" CONTENT="CV.h">\r
- <TITLE>CAMShift Demo</TITLE>\r
- <!-- Formatted and proofed in GNU Emacs by: Michael Chu, 00.06.05 -->\r
- </HEAD>\r
- <BODY TEXT="#000000" BGCOLOR="#FFFFFF" LINK="#0000FF" VLINK="#800080" ALINK="#FF0000">\r
- <A NAME="Top">\r
- <!-- <CENTER>\r
- <BLOCKQUOTE>\r
- <table border = 3 WIDTH="70%" >\r
- <TR><TD>\r
- -->\r
- <CENTER><TABLE CELLSPACING=0 CELLPADDING=5 WIDTH="90%" BGCOLOR="#6A9BED" NOSAVE >\r
- <TR NOSAVE>\r
- <TD NOSAVE>\r
- <CENTER><I><FONT COLOR="#000000"><FONT SIZE=+4>CAMSHIFT\r
- DEMO</FONT></FONT></I></CENTER>\r
-\r
- <CENTER><I><FONT COLOR="#000000"><FONT SIZE=+4>Face and Color Tracking</FONT></FONT></I></CENTER>\r
- </TD>\r
- </TR>\r
- </TABLE></CENTER>\r
-\r
- <H1>\r
- <!--\r
- </TD></TR>\r
- </table>\r
- </BLOCKQUOTE>\r
- </center>\r
- --></H1>\r
-\r
- <br>\r
- <H1>\r
- Description</H1>\r
- This application demonstrates a fast, simple color tracking algorithm that\r
- can be used to track faces, hands etc described in <B><A HREF="../../papers/camshift.pdf">camshift.pdf</A></B>.\r
- The CAMSHIFT algorithm is a modification of the Meanshift algorithm which\r
- is a robust statistical method of finding the mode (top) of a probability\r
- distribution. Both CAMSHIFT and Meanshift algorithms exist in the library.\r
- While it is a very fast and simple method of tracking, because CAMSHIFT\r
- tracks the center and size of the probability distribution of an object,\r
- it is only as good as the probability distribution that you produce for\r
- the object. Typically the probability distribution is derived from\r
- color via a histogram, although it could be produced from correlation,\r
- recognition scores or bolstered by frame differencing or motion detection\r
- schemes, or joint probabilities of different colors/motions etc.\r
- \r
- <br><br>\r
- In this application, we use only the most simplistic approach:\r
- A 1-D Hue histogram is sampled from the object in an HSV color space version\r
- of the image. To produce the probability image to track, histogram\r
- "back projection" (we replace image pixels by their histogram hue value)\r
- is used. The raw video image and backprojection is shown in the following\r
- image:<br><br>\r
- <CENTER><IMG SRC="faceNback.png" HEIGHT=130 WIDTH=347></CENTER>\r
-\r
- <CENTER><B>Figure 1</B></CENTER>\r
-\r
- <br>\r
- Since we use such a simple method in this application, you must tune\r
- the parameters for your camera and the lighting in the image as described\r
- below.\r
- <BR><BR>\r
- Note that this filter requires that you have DirectShow 6.0 or greater installed.\r
- DirectShow 6.0 for NT4.0, DirectShow 7.0 for Win 98 or Win2000). \r
- You will also need a video camera source filter which should\r
- be supplied by your USB camera vendor. <!-- except most don't support NT4.0. -->\r
-<!-- %%Some friendly advice from Uncle Gary?? -->\r
-<!-- Win2000 is your best bet for development, for just running the filter graph, Win98 will do.. An example filter graph is shown below (with the calibration filter processing a video source from a Matrox Meteor II card that we wrote ourselves) and then passing the output to a video renderer. -->\r
-\r
- Microsoft* Windows* 2000 may be the most suitable operating system\r
- for development since many vendors do not supply filters for\r
- Windows NT 4.0. However, Windows 98 will be adequate for just\r
- running the filter graph. \r
- <br>\r
- <!-- <HR WIDTH="100%"> -->\r
- <H2>\r
- Controls</H2>\r
- \r
- <CENTER><IMG SRC="csmshiftTrack_labeled.JPG" HEIGHT=319 WIDTH=381></CENTER>\r
-\r
- <CENTER><IMG SRC="camshiftThresh_labeled.JPG" HEIGHT=138 WIDTH=855></CENTER>\r
-\r
- <CENTER><B>Figure 2</B></CENTER>\r
- \r
- <!-- <H2>Buttons</H2> -->\r
- <B><FONT SIZE=+2>[1]</FONT></B> Select input from USB video camera.\r
- <BR><B><FONT SIZE=+2>[2]</FONT></B> Select input from "bouncing ball" simulation.\r
- <BR><B><FONT SIZE=+2>[3]</FONT></B> Select input from an AVI movie file.\r
- <BR><B><FONT SIZE=+2>[4]</FONT></B> Stop and start AVI movie file if [3]\r
- is selected.\r
- <BR><B><FONT SIZE=+2>[5]</FONT></B> Threshold the histogram bins to clean\r
- noise.\r
- <BR><B><FONT SIZE=+2>[6]</FONT></B> Select the number of histogram bins\r
- to use in the 1-D Hue histogram.\r
- <BR><B><FONT SIZE=+2>[7]</FONT></B> Minimum color saturation to accept\r
- threshold on Hue pixels.\r
- <BR><B><FONT SIZE=+2>[8]</FONT></B> Minimum brightness to accept threshold\r
- on Hue pixels.\r
- <BR><B><FONT SIZE=+2>[9]</FONT></B> Maximum brightness to accept threshold\r
- on Hue pixels.\r
- <BR><B><FONT SIZE=+2>[10]</FONT></B> View (top) Raw video; (middle) Histogram\r
- backprojection; (bottom) Histogram bins.\r
- <BR><B><FONT SIZE=+2>[11]</FONT></B> Select display window size -- on slower\r
- machines, a smaller display window will run faster.\r
- <BR>\r
- <H2>\r
- How to</H2>\r
- <h3>Track a colored object (face, ball etc):</h3>\r
- (<B>a</B>) Reset thresholds: Set Threshold [5] to zero; Saturation\r
- [7] far left; Minimum brightness [8] far left; Maximum brightness [9] far\r
- right.\r
- <BR>(<B>b</B>) Click video [1], bouncing ball [2] or AVI [3] to select\r
- your input source.\r
- <BR>(<B>c</B>) Check the raw video display [10] button.\r
- <BR>(<B>d</B>) Set the number of bins [6] you want in your histogram (32-100\r
- is a good number).\r
- <BR>(<B>e</B>) Use your mouse to draw a sample box over the colored object\r
- that you want to track.\r
- <h3>...Tuning...</h3>\r
- (<B>f</B>) Check the histogram "Back Project" [10] button. You\r
- want the object to be well segmented as in figure 1 above. To get\r
- things well segmented, adjust things in the order described below:\r
- <BR>(<B>g</B>) Increase (move the slider to the right) the mimimum acceptable\r
- saturation threshold [7]. Make sure you allow a little time to see\r
- how your adjustment is working. Often moving it about 25% of the way to\r
- the right is sufficient.\r
- <BR>(<B>h</B>) If (g) doesn't do the whole job, then increase the minimum\r
- acceptable brightness threshold [8] to the right allowing enough time to\r
- see the effect at each position. Usually 20-50% of the way to the right\r
- works.\r
- <h3>...not working...</h3>\r
- (<B>i</B>) If (g) & (h) do not give you a well segmented object,\r
- you probably need to change your camera settings and/or lighting.\r
- Histogram threshold [5] can have a marginal effect on edge regions.\r
- Decreasing (move to the left) high brightness threshold [9] can diminish\r
- edges and glaring white regions in the video image, usually moving it no\r
- more than 10% of the way to the left. Setting more histogram bins\r
- [6] sometimes helps.\r
- <BR>\r
- <BR>\r
-\r
- <hr noshade size=2 width=100%>\r
-\r
- </BODY>\r
-</HTML>\r
projects / opencv / blobdiff