Boost::Python with OpenCV

In my previous post Configuring Boost::Python and Hello Boost, I have shown how to configure Boost::Python. I have managed to use OpenCV C++ with Python, using Boost::Python. I have been able to load and show an image successfully.

OpenCV C++:

First, write a simple function which will load and if required show an image using opencv. I have named it opencvtest.cpp

#include<opencv2/highgui/highgui.hpp>
using namespace cv;
using namespace std;

int load(string filename)
{

    Mat img = imread(filename,CV_LOAD_IMAGE_COLOR);
    //imshow("opencvtest",img);
    //waitKey(0);
    return 1;
}

Wrapping:

Wrap the above function using Boost::Python.

#include<boost/python.hpp>
using namespace boost::python;
BOOST_PYTHON_MODULE(opencvtest)
{
    def("load",load);
}

Boost::Python wrapping is successfully done. I tried using bjam, but there were many complications. So, I am sticking to make.

Makefile:

PYTHON_VERSION = 2.7
PYTHON_INCLUDE = /usr/include/python$(PYTHON_VERSION)

# location of the Boost Python include files and library

BOOST_INC = /usr/include
BOOST_LIB = /usr/lib

# your filename here
TARGET = opencvtest

$(TARGET).so: $(TARGET).o
	g++ -shared -Wl,--export-dynamic $(TARGET).o -L$(BOOST_LIB) -lboost_python -L/usr/lib/python$(PYTHON_VERSION)/config -lpython$(PYTHON_VERSION) -o $(TARGET).so `pkg-config --libs opencv` `pkg-config --cflags opencv`

$(TARGET).o: $(TARGET).cpp
	g++ -I$(PYTHON_INCLUDE) -I$(BOOST_INC) -fPIC -c $(TARGET).cpp

Notice `pkg-config` added in the Makefile.
 

$ make

 
opencvtest.o and opencvtest.so binaries will be created.

Now, use the wrapper with Python.

Python:

import time
def timetest():
    image = "filename.jpg"
    t0 = time.time()
    from cv2.cv import LoadImage
    img = LoadImage(image)
    t1 = time.time()
    from opencvtest import load
    img2 = load(image)
    t2 = time.time()
    print "Time taken to load image using cv2.cv module:",
    print t1-t0
    print "Time taken to load image using boost wrapper:",
    print t2-t1

timetest()

Notice the significant difference between run time span of both the methods.
 
Here’s the complete gist.
 
Even I couldn’t believe that I could easily wrap OpenCV functions using Boost::Python. Now, I am planning to use Boost::Python to wrap OpenTLD and make a python port as well as wrap some important and time consuming iterations, functions using boost so that run-time speed of SimpleCV can be increased.
 
P.S. Going to Hyderabad to meet my brother. This is going to be exciting.

About these ads