MosseTracker

Description

MOSSE tracker is used to track a single object on the frame sequence.

Algorithm is based on the paper: "Visual Object Tracking using Adaptive Correlation Filters" by David S. Bolme, J. Ross Beveridge, Bruce A. Draper, Yui Man Lui, Computer Science Department, Colorado State University. Original Fast Fourier Transform algorithm was written by Miroslav Voinarovsky (distributed as freeware open source) with insignificant modifications by me. The usage of fftw (single precision, distributed via GNU GPL) can be defined instead of FFT by Miroslav Voinarovsky.

Compiled via MS Visual Studio 2015. Uses std and Windows.h. Should be slightly modified for Lin-support. Uses OpenCV to output test results, OpenCV distribs are not included.

Algorithm is singlethreaded and provides no thread safety. Thread safety should be provided by the control application. FPS is calculating via QueryPerformanceCounter. FPS is about 7000 for the 64x64 rectangle of interest (with fftw) or about 1600 without it (Intel Core i5, 3.2GHz). Warning: FPS calculated in TestFromFile can be inaccurate. It is better to estimate FPS via TestOneFrame. Suggestions for more accurate FPS counting are welcome.

Structure

1. MosseTracker project (dll)

2. MosseTracker module 1. MosseTracker.h 2. MosseTracker.cpp - contains main tracker functions - initialization and onFrame 3. MosseTracker_Misc.cpp - contains misc tracker functions 4. MosseTracker_Debug.cpp - contains some debug functions

3. Export section 1. Mosse_Export.h - contains declaration of exported functions 2. Mosse_Export.cpp - contains definition of exported functions, controls MosseTracker objects lifetime

4. FFT - Fast Fourier Transform 1. Fft.h 2. Fft.cpp 3. Fft_arrays.h - contains large magic arrays used for FFT calculations

5. MosseSamples (exe)

6. Samples 1. SampleFromFile - runs MosseTracker sample with the images loaded from files 2. SampleOneFrame - runs sample with only one frame (and one for the initialization). It processes this frame 5000 times and counts the total time 3. Sample4Trackers - runs sample with 4 trackers purely for academic purposes (that is, for entertainment) 4. SampleSwarm - runs sample with a lot of trackers (e.g. 100). New position is taken as the average of their offsets

7. fftw distrib

8. fftw3.h

9. libfftw3f-3.dll

10. libfftw3f-3.lib

11. Test image sequences

<br> Anton Egorov, 2016<br> dohxehapo@gmail.com