Changeset 645

Timestamp:

01/31/10 23:31:20 (2 years ago)

Author:

mosttw

Message:

Status report, corrections

Location:

branches/2010-image-rec/docs

Files:

• Makefile (modified) (2 diffs)
• figures (added)
• figures/cam-000080-hsv-sobel-h.png (copied) (copied from branches/2010-image-rec/images/elphel/cam-000080.jpeg-hsv-sobel-h.png)
• project-plan.tex (modified) (4 diffs)

branches/2010-image-rec/docs/Makefile

@@ -3 +3 @@
 default: all
 
-all: lit-review.pdf
+project-plan.pdf: figures/cam-000080-hsv-sobel-h.png figures/hist-test.tex figures/pygments-declarations.tex
+
+all: lit-review.pdf project-plan.pdf
 
 %.pdf: %.tex
@@ -10 +12 @@
         pdflatex $(TEXFLAGS) $<
         pdflatex $(TEXFLAGS) $<
-        rm -f $*.aux $*.bbl $*.blg $*.brf $*.log report.out $*.tex~
+        rm -f $*.aux $*.bbl $*.blg $*.brf $*.log $*.out $*.tex~
 
 

branches/2010-image-rec/docs/project-plan.tex

@@ -1 @@
-\documentclass[11pt,letterpaper,oneside]{article}
+\documentclass[10pt,letterpaper,oneside]{article}
 \usepackage{graphicx}
 \usepackage{fullpage}
@@ -14 +14 @@
 
 \begin{document}
+\input{figures/pygments-declarations.tex}
 
 \begin{flushright}
@@ -23 +24 @@
 %\begin{multicols}{2}
 
-\subsection*{Report}
+\subsection*{Status Report}
 
+At this point we have gathered a number of test images, examined the suitability of texture for differentiating grass and obstacles, and completed a simple color histogram-based prototype implementation in Matlab.  The test images we have gathered were taken at previous IGVC competitions, both using a handheld digital camera and the Elphel camera actually used on the robot.  The different color qualities of these image sets will help us test that our algorithm can handle different lighting conditions and color tones.  A small script has been written to apply a Sobel filter to all of these images, allowing us to see the texture of the images.  This has been done with images in the RGB and HSV colorspaces.  We determined that the Sobel of the hue band is most suitable for this, as shown in Figure \ref{fig:h-sobel}.
+
+\begin{figure}[htb]
+        \centering
+                \includegraphics[width=0.5\textwidth]{figures/cam-000080-hsv-sobel-h.png}
+        \caption{The hue band of a test image with the Sobel filter applied.}
+        \label{fig:h-sobel}
+\end{figure}
+
+Finally, a preliminary Matlab implementation of the algorithm presented by Ulrich and Nourbakhsh has been written.  This is purely based on color (in the HSV space) at this point.  The next step will be to integrate a measure of texture (likely a regional average of the Sobelized image) into the algorithm.  The full text of this script is presented in Appendix \ref{appendix:hist-test}.
 
 \subsection*{Iterative enhancement plan}
 \begin{enumerate}
 \item
-For our easy goal we hope to be able to reliably classify pixels within a single static image as either obstacle, line, or grass. We will implement an algorithm for classifying pixels as line or obstacle and grass based on a `safe zone.' It then classifies obstacles and lines or grass using texture. Combining these classifiers should give reliably sepperate between our three classifications.
+For our easy goal we hope to be able to reliably classify pixels within a single static image as either obstacle, line, or grass. We will implement an algorithm for classifying pixels as line or obstacle and grass based on an assumed ``safe zone'' in front of the robot.  Combining color and texture qualities should allow these classifiers to reliably identify members of the three classes.
 \item
-Our more difficult goal will be to take these classified pixels and make them into useful information on the robots environment. For the obstacle pixels we will identify objects full objects. Then use a perspective transformation or approximation to find their distance and location relative to the robot. For the lines we will use a Hough transform to find the equations of visible, broken, and partially hidden lines. Lines will also require an as yet determined approximation for curved regions.
+Our more difficult goal will be to take these classified pixels and derive useful information about the robot's environment from them. Real-world obstacles will be identified based on clusters of obstacle pixels. Through use of a perspective transformation or an approximation thereof, their real-world distance and location relative to the robot will be recovered.  For the lines we will use a Hough transform to find the equations of visible, broken, and partially hidden lines. Lines will also require an as yet determined approximation for curved regions.
 \item
-Our stretch goal will be to take all our work, performed in Matlab, and port it into OpenCV, with the goal of using on an actual robot in real time.
+Our stretch goal will be to take all our work, performed in Matlab, and port it to OpenCV, to be used on an actual robot in real time.
 \end{enumerate}
 
@@ -42 +53 @@
 %\end{multicols}
 
+\pagebreak
+
+\appendix
+\section{\texttt{hist\_test.m}}
+\label{appendix:hist-test}
+
+\input{figures/hist-test.tex}
+
 \end{document}