ferns-demo
----------
This software demonstrates Fern based keypoint matching for planar
object detection. It also includes template based tracking code for
applications that demand even faster operation.
Homepage: http://cvlab.epfl.ch/software/ferns/index.php
Usage:
------
Type
$ make
to compile the code and generate the executable 'ferns-demo'.
The code depends on the OpenCV library, so you should first install
and make sure that pkg-config can locate it. To test just type
$ pkg-config opencv --libs --cflags
and you should get a list of compiler flags.
The executable by default uses 'model.bmp' and 'model.bmp.roi' files
to train a Fern based classifier and save it to
'model.bmp.detector_data'. It then tries to open a webcam and start
detecting the object. You can use the '-v' flag to specify a video
file as image source. So to try detection on a video, type
$ ./ferns-demo -v mousepad.mp4
You can also use a sequence of images to test detection,
$ ./ferns-demo -s 'image%04.jpg'
To change the model object just take a photo of a planar object and
save it. You can specify a region of interest by writing a roi file
the same name as your model file + '.roi' extension. You can specify
the name of the image using the '-m' option,
$ ./ferns-demo -m book.jpg
which will use 'book.jpg' and 'book.roi' files to generate
'book.jpg.detector_data'
Type
$ ./ferns_demo -h
to see a list of command line options.
To change parameters for training just edit 'main.cc' and see the
header file for 'planar_pattern_detector_builder.h'.
Template Based Tracking:
------------------------
Tracking can be done much faster than detection but is less reliable
and needs initialization and reinitialization in case of failure. So
the executable has different modes of operation that can be switched
by pressing the number keys (0-3).
0: Detect when tracking fails or for initialization then track.
1: Track only
2: Detect only (DEFAULT)
3: Detect + track in every frame
The number keys 4&5 can be used to turn on/off the recognized and
detected keypoints, respectively.
By default the tracker is trained using the same model file as the
detector and the tracker data is saved in a file having the same name
as the model file + '.tracker_data' extension.
Windows:
--------
There is no automated way to compile the code under Windows. It should
be relatively easy to write one, since the source code depends only on
OpenCV (which has a Windows version) and a decent C++ implementation.