fragtrack_console_app
所属分类:OpenCV
开发工具:Visual C++
文件大小:26KB
下载次数:61
上传日期:2011-04-21 11:08:05
上 传 者:
amciyo
说明: 遮挡情况下最稳定有效的物体跟踪算法 fragtracker -- adam, cvpr 06
(The most stable and effective under occlusion in object tracking algorithm fragtracker- adam, cvpr 06)
文件列表:
fragtrack_console_app\emd.cpp (19841, 2005-10-09)
fragtrack_console_app\emd.h (1510, 2006-08-02)
fragtrack_console_app\Fragments_Tracker.cpp (26790, 2008-01-21)
fragtrack_console_app\Fragments_Tracker.h (4902, 2008-01-21)
fragtrack_console_app\fragtrack_console_app.sln (1687, 2007-03-18)
fragtrack_console_app\fragtrack_console_app.suo (15872, 2008-01-21)
fragtrack_console_app\fragtrack_console_app.vcproj (3857, 2007-03-18)
fragtrack_console_app\fragtrack_envelope.cpp (5424, 2008-01-21)
fragtrack_console_app\FragTrack_log.txt (383, 2008-01-21)
fragtrack_console_app\setup.txt (66, 2008-01-21)
fragtrack_console_app\setup_face_sequence.txt (66, 2008-01-21)
fragtrack_console_app\setup_woman_sequence.txt (69, 2008-01-21)
fragtrack_console_app (0, 2008-01-21)
FragTrack - Fragments-based Tracking Code
-----------------------------------------
By: Amit Adam
amita@cs.technion.ac.il
www.cs.technion.ac.il/~amita
Date: November 18th, 2007
-----------------------------------------
General
-------
This distribution contains the source code for a fragments-based tracker.
It is written in C++ and uses the OpenCV library.
What's in the package
---------------------
1. Fragments_Tracker.h,cpp - the tracker object code
2. fragtrack_envelope.cpp - an envelope for running the tracker on an image sequence
3. emd.h,cpp - code for comparing two histograms using Earth Mover's Distance -
courtesy of Yossi Rubner
4. A Visual Studio solution for building the project.
5. Sample setup files for two image sequence. The sequences may be found in my
homepage.
6. Sample log file.
Usage
-----
1. Build the executable - a console application (tested only in "release configuration")
2. Prepare a setup file called "setup.txt" and place it in the same directory
as the executable file.
3. Run the executable.
4. The following output should be obtained:
- during the run - an OpenCV window with the tracking results
- a log file called "FragTrack_log.txt" containing the tracking results
- two images "initial_temlate.jpg" and "initial_target.jpg" showing the initial template
Format of setup file
--------------------
The setup file is a text file containing 7 lines in the following format:
F:\\amita\\data\\face_sequence\\ % line 1 - path and file name prefix
1 % line 2 - first image in sequence
890 % line 3 - last image in sequence
75 120 220 235 % line 4 - target position in first image
7 % line 5 - search window half size
16 % line 6 - number of bins in histogram
3 % line 7 - choice of metric for comparing histograms
(do not include the comments in the setup file)
Here are some details:
The first 3 lines specify where to find the input sequence - in the above example the sequence is
F:\amita\data\face_sequence\ 1.jpg, F:\amita\data\face_sequence\2.jpg, ..., F:\amita\data\face_sequence\890.jpg
Note: no spaces are allowed in path or file name
Line number 4 gives the top-left and bottom-right corners of the target position in the first frame.
The y-coordinate (row) is given first: tl_y tl_x br_y br_x
Line number 5 specifies the search radius around the position in previous frame (in pixels).
The algorithm is based on gray-scale intensity histograms. Line number 6 specifies the number of bins
in the histograms.
For comparing two histograms the algorithm currently uses one of three options. Line 7 specifies which option:
1 means chi-square metric, 2 means EMD metric, 3 means a variation of the Kolmogorov-Smirnov statistic. The EMD
is a cross-bin metric in contrast with standard bin-to-bin metrics such as Chi square.
For one dimensional data option 3 is a much faster equivalent to the EMD metric. Option 3 should be
your default choice. You can see the advantage option 3 has over option 1 on the "woman" sequence for example.
Example setup files and sequences
---------------------------------
Two example setup files are contained in the distribution. The corresponding sequences are available from
www.cs.technion.ac.il/~amita
together with a file containing the ground truth for these sequences.
Feedback
--------
Feedback (both positive and negative) is most welcome. Please email to amita@cs.technion.ac.il
Acknowledgement
---------------
Thanks to Yossi Rubner for his EMD code and for permission to redistribute it with this package.
Reference
---------
Amit Adam, Ehud Rivlin, Ilan Shimshoni: Robust Fragments-based Tracking using the Integral Histogram.
Proc. CVPR 2006, pp. 7***-805
近期下载者:
相关文件:
收藏者: