http:^^www.cs.cornell.edu^info^people^hejik^cs631^paper.html
来自「This data set contains WWW-pages collect」· HTML 代码 · 共 431 行 · 第 1/2 页
HTML
431 行
MIME-Version: 1.0
Server: CERN/3.0
Date: Sunday, 01-Dec-96 19:45:11 GMT
Content-Type: text/html
Content-Length: 15559
Last-Modified: Saturday, 09-Dec-95 04:57:20 GMT
<HTML><TITLE>ORWELL: Removal of Tracked Objects in Digital Video</TITLE><body bgcolor="#ffffff" text="#000000" link="#60a0c0" vlink="#60a0c0" ><BODY><H1><P ALIGN=CENTER>ORWELL: Removal of Tracked Objects in Digital Video</A></H1><P ALIGN=CENTER><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><!WA0><a href="http://www.cs.cornell.edu/Info/People/ahong/home.html">Alfred Hong</a>,<!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><!WA1><a href="http://www.cs.cornell.edu/Info/People/hejik/hejik.html">Heji Kim</a>,<!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><!WA2><a href="http://www.cs.cornell.edu/Info/People/lhwang/lhwang.html">Lin-hsian Wang</a><p><P ALIGN=CENTER>Department of Computer Science<BR>324 Upson Hall<BR>Cornell University<BR>Ithaca, NY 14853-7501 US<BR><BR><p><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><!WA3><a href="http://www.cs.cornell.edu/Info/Projects/zeno/rivl/rivl.html"><P ALIGN=CENTER>http://www.cs.cornell.edu/Info/Projects/zeno/rivl/rivl.html</a><p><hr><H2><A NAME="toc">Table of Contents</A></H2><UL><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><!WA4><A HREF="http://www.cs.cornell.edu/Info/People/hejik/cs631/paper.html#HDR2"><B>1. INTRODUCTION</B></A><BR><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><!WA5><A HREF="http://www.cs.cornell.edu/Info/People/hejik/cs631/paper.html#HDR3"><B>2. BACKGROUND INFORMATION</B></A><BR><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><!WA6><A HREF="http://www.cs.cornell.edu/Info/People/hejik/cs631/paper.html#HDR4"><B>3. SPECIFICS</B></A> <UL> <!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><!WA7><A HREF="http://www.cs.cornell.edu/Info/People/hejik/cs631/paper.html#HDR5">3. 1 Object-Tracking: Hausdorff Tracker</A> <BR> <!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><!WA8><A HREF="http://www.cs.cornell.edu/Info/People/hejik/cs631/paper.html#HDR6">3. 2 Background-Reconstruction</A> <BR> <!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><!WA9><A HREF="http://www.cs.cornell.edu/Info/People/hejik/cs631/paper.html#HDR7">3. 3 Object-Segmentation</A> </UL><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><!WA10><A HREF="http://www.cs.cornell.edu/Info/People/hejik/cs631/paper.html#HDR8"><B>4. EVALUATION</B></A><BR><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><!WA11><A HREF="http://www.cs.cornell.edu/Info/People/hejik/cs631/paper.html#HDR9"><B>5. RELATED WORK AND EXTENSIONS</B></A><BR><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><!WA12><A HREF="http://www.cs.cornell.edu/Info/People/hejik/cs631/paper.html#HDR10"><B>6. REFERENCES</B></A></UL><p><HR><H2><A NAME="HDR2">1. Introduction</A></H2>Object tracking in a sequence of images can provide a base for a multitude of digital videoprocessing applications such as removal of object in the scene.Although numerous video-processing editors are available,object-tracking and removal (OTR) is mostly a manual process. Using the existing object-tracking feature in <!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><!WA13><A HREF="http://cs.cornell.edu/Info/Projects/zeno/rivl/rivl.html">RiVL</A>, we implement a semi-automated application that allows the user to specify and remove an object, then reconstruct the background to result in a new video sequence. Our work primarily focuses on algorithms for the domain of stationary backgrounds with a single moving object. <p>In addition to OTR, we also extend this work to segment the tracked-object from the background;we can use the resulting segmentation for a variety of video processing effects such asoverlaying the tracked-object on top of different sequence. The resulting application is an ideal test bed for experimenting withvarious OTR and segmentation algorithms.We reconstruct the background and use different techniques for segmentation as illustrated in the diagram below.</A><P ALIGN=CENTER><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><!WA14><img src="http://www.cs.cornell.edu/Info/People/hejik/cs631/digram.gif"><B><A NAME="REF69077"><BR>Figure 1:</B> Orwell OTR and Segmentation Overview</P>The rest of the paper is organized as follows.<UL><LI>Section 2 provides some background on RiVL and the Hausdorff tracking algorithm<LI>Section 3 discusses vision algorithms we employed to our ends <LI>Section 4 reviews the work we have done<LI>Section 5 concludes with current status and future research directions.</UL><p>[Key words: <i>object-tracking, Hausdorff distance, object-removal, segmentation, background reconstruction, image filtering</i>]<p><H5><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><!WA15><A HREF="#toc"><-- Table of Contents</A></H5><br><H2><A NAME="HDR3">2. Background Information</A></H2><H3>RiVL</H3>RiVL is a resolution independent video language whichhas video and audio as first class data types. Jonathan Schwartz has implemented RiVL as a Tcl/Tkextension for multimedia processing. The high level operators used in RiVL areindependent of video format and resolution and provides the necessary infrastructure to test our ideas.<H3>RiVL_GenC</H3>RiVL_GenC generates the C code for RiVL functions that need to perform low level image processing routines that are not already included in the RiVL library. The implementations of median and mean filters use the functions generated by RiVL_GenC for pixel-level computations.<H3><A NAME="Idontknow"> Hausdorff Tracker</A></H3>The Hausdorff tracker is a feature-based object tracking system for a continuous sequence of images.The model of the tracked object is represented by a binary edge-mapproduced by applying a Canny edge operator over a smoothed versionof the gray-level image of the input image. Taking advantage of the fact that the motionof the object is roughly an affine transformation between any two consecutive frames, the algorithm matches all the possible translations and scales of the model to a specified search window (shown as a red dotted boxin Figure 2). Generally, the best match has the most points which overlap with a transformation of the model.Since we use this best match as a model for the next image, once the tracker begins to wander, the results can deteriorate quickly.<p> <c><P ALIGN=CENTER><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><!WA16><img src="http://www.cs.cornell.edu/Info/People/hejik/cs631/hausdorff.gif"><B><A NAME="REF69077"><BR>Figure 2:</B> Hausdorff tracking algorithm explained</P> </c><p><p><H5><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><!WA17><A HREF="#HDR3"><-- Background Information</A></H5><H5><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><!WA18><A HREF="#toc"><-- Table of Contents</A></H5><H2><A NAME="HDR4">3. Specifics</A></H2>This section discusses the implementation of the algorithms accessible from the Orwell Editor. The first subsection discusses object-tracking. The second subsection discusses the background reconstruction algorithmsassuming a stationary camera, and the third subsection discusses the segmentation algorithms. <p><H3><A NAME="HDR5">3. 1 Object-Tracking: Hausdorff Tracker</A></H3>The tracker in RiVL returns scale and translation coordinates for each image. Performance of the tracker depends on setting the correct parameters for the search, i.e. size of the search window, scaling factors, and the forward and backward distance which limits the allowed dissimilarity in a match. We must make a trade-off between the laxness of the constraints and the processing time required totrack the object.The Hausdorff tracker also works better for larger images.</A><p><H5><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><!WA19><A HREF="#HDR3"><-- Background Information</A></H5><H3><A NAME="HDR6">3. 2 Background Reconstruction</A></H3>We need the background to replace the tracked objectfrom the original sequence and to possibly segment the object.We experiment with three different approaches tobackground reconstruction: a temporal median filter, a temporal mean filter,and a physical space search.<p></p><p> The first two approaches for the background reconstruction are temporal mean filter and the temporal median filter.</p><H3><A NAME="HDR6">Temporal Mean Filter (TMEF)</A></H3><p> The TMEF technique computes the mean pixel value by taking the arithmetic average of the whole frame sequence and assigns this result to the pixels in the background frame.This technique averages out the tracked object in the scene with a possible blurringeffect. We implement this filter by averaging each of the RGB values independently.</p> <H3><A NAME="HDR6">Temporal Median Filter (TMDF)</A></H3>⌨️ 快捷键说明
复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?