Haar Classifier

1. Preparing the following files
   
   • ..\OpenCV\samples\c\convert_cascade.exe
   • ..\OpenCV\bin\createsamples.exe
   • ..\OpenCV\bin\haartraining.exe
     
     
2. Preparing positive and negative samples
   
   • Negative( "..\negative\" )
     • dir /b/s > neg.txt
     • replacing "..\" to ""
   • Positive( "..\positive\" )
     
     • dir /b/s > pos.txt
     • replacing "..\" to ""
     • adding "NUM X Y WIDTH HEIGHT" to each end of line
       e.g. positive\a.png 1 0 0
       
       
3. Moving "pos.txt" and "neg.txt" to "..\"
   
   
4. Creating the positive vector
   createsamples -info pos.txt -vec pos.vec -num POSTOTAL -w WIDTH -h HEIGHT
   e.g. createsamples -info pos.txt -vec pos.vec -num 35784 -w 32 -h 24
   
   
5. (Optional) Checking samples
   createsamples -vec pos.vec -w WIDTH -h HEIGHT
   e.g. createsamples -vec pos.vec -w 32 -h 24
   
   
6. Training
   haartraining -data data\cascade -vec pos.vec -bg neg.txt -npos POSTOTAL -nneg NEGTOTAL -nstages STAGE -mem MEMORY -mode ALL -w WIDTH –h HEIGHT
   e.g. haartraining -data data\cascade -vec pos.vec -bg neg.txt -npos 35784 -nneg 3240 -nstages 30 -mem 1000 -mode ALL -w 32 –h 24
   
   
7. Generating a xml file
   convert_cascade --size="x" PATH CLASSIFIER
   e.g. convert_cascade --size="32x24" \data\cascade\ haarcascade.xml
   

[Reference]

• Tutorial: OpenCV haartraining
  

LaTex Memo

Command
[Create PDF]
pdflatex FILENAME



Basic example
\documentclass[parameter]{format}
\usepackage{name}
\begin{document}
\title{string}
\author{string}
\section{string}
\subsection{string}
\end{document}

[\documentclass]
parameter
e.g. 10pt, a4paper
format
e.g. article, report, CLS file name

[\usepackage]
name
e.g. subfigure, graphicx



Image insertion
\begin{figure}
\begin{center}
\includegraphics[format]{name}
\caption{string}
\label{string}
\end{center}
\end{figure}

[\begin{center}]
置中

[\includegraphics]
format
e.g: width=0.4\textwidth
name
e.g: name.png, name.jpg

[\caption{string}]
圖片的標題

[\label{string}]
\ref{string}用



Subfigure
\subfigure[string]
{
　　\label{string}
　　\includegraphics[format]{name}
}
\subfigure[string]
{
　　\label{string}
　　\includegraphics[format]{name}
}

[\subfigure]
string
子圖標題



Bibliography
\begin{thebibliography}{num}
\bibitem{tag}
\end{thebibliography}

[\begin{thebibliography}]
num
起始數字

[\bibitem]
tag
\cite{tag}用

Vision for a Smart Kiosk

Vision for a Smart Kiosk
James M. Rehg, Maria Loughlin, Keith Waters



The kiosk interface supports public interaction with multiple users.

[Obtained Information]

• three dimensional location
• body language
• facial expressions

[Modules]

• Motion blob detection
  
• Color tracking
  a color histogram model of each user's shirt
  
• Stereo triangulation
• DECface
• Behavior
  

[DECface Agent]

• speak an arbitrary piece of text at a specific speech rate in one of eight voices from one of eight faces
• the creation of simple facial expressions under control of a facial muscle model
• simple head and eye rotation

[Taxonomy of vision tasks]

	Features	Attributes	Behaviors
Distant (Dist)	whole body	position	monitor
Midrange (Mid)	head and torso	orientation	entice
Proximate (Prox)	face / hands	expression	communicate

[Experiments]

• 推測position的正確性
• tracking和behvior模組的測試
  

[Conclusion]

• Color is a valuable feature for tracking people in real-time.

Median filter

[Step][Example]

1. Decide a window with odd samples.
2. Calculate the median of these samples.

[Purpose]

• 消除雜訊

Funny about Molly

Solving a learning system (Ax=b) by iterative methods

Jacobi Method
for i = 1 : n
　x[ i ]^k+1 = ( b[ i ]
　　　　　　　 - SIGMA(j = 1 : i-1)( a[ i ][ j ] * x[ j ]^k )
　　　　　　　 - SIGMA(j = i+1 : n)( a[ i ][ j ] * x[ j ]^k ) )
　　　　　　/ a[ i ][ i ]
end

[Convergence]
M = diag( diag( A ) )　N = - ( A - M )
若M^-1*N的eigenvalue皆在-1~1之間的話，則其會收斂。



Gauss-Seidal Method
for i = 1 : n
　x[ i ]^k+1 = ( b[ i ]
　　　　　　　 - SIGMA(j = 1 : i-1)( a[ i ][ j ] * x[ j ]^k+1 )
　　　　　　　 - SIGMA(j = i+1 : n)( a[ i ][ j ] * x[ j ]^k ) )
　　　　　　/ a[ i ][ i ]
end



Steepest Descent Method
x(0) = initial guess
r(0) = b - A * x(0)
k = 0
while r^k != 0
　k = k + 1
　alpha(k) = ( (r^k-1)' * r^k-1 ) / ( (r^k-1)' * A * r^k-1 )
　x^k = x^k-1 - alpha^k * r^k-1
　r^k = b - A * x^k
end



Conjugate Gradient Method
k = 0
r(0) = b - A * x(0)
while r^k != 0
　k = k + 1
　if k = 1
　　p(1) = r(0)
　else
　　beta^k = ( (r^k-1)' * r^k-1 ) / ( (r^k-2)' * r^k-2 )
　　p^k = r^k-1 + beta^k * p^k-1
　end
　alpha^k = ( (r^k-1)' * r^k-1 ) / ( (p^k)' * A * p^k )
　x^k = x^k-1 + alpha^k * p^k
　r^k = r^k-1 - alpha^k * A * p^k
end
x = x^k



[Reference]

• Gene H. Golub and Charles Van Loan, Matrix computation, 1996.
  [Part 1] [Part 2]

Evaluation

		Actual condition
		True	False
Test result	True	True Positive (TP)	False Positive (FP)
	False	False Negative (FN)	True Negative (TN)

Accuracy = ( TN + TP ) / ( TP + FN + FP + TN )

Precision = TP / ( TP + FP )

Recall = TP / ( TP + FN )