Abstract:
There are many software packages available in Python in computer vision. Nevertheless, in
order to deal with more low level faster algorithms and embedded systems, C++ is preferable.
Installing and configuring an OpenCV environment may not be included in a teaching module's
learning outcomes. This review involves finding the feasibility of properly executing OpenCVbased
computer vision programmes in C++ in a Windows environment with minimal preconfiguration.
The OpenCV libraries were built using the C++ source, and the generated
libraries and header files were attached to a Visual Studio project which can be easily distributed
among others. Building and configuring OpenCV may need an experienced user, but a simple
tutorial or guideline is enough to train and use. Bilateral filtering and edge, face, HOG detection
were the chosen algorithms that were run with parallel processing in CPU and with and without
OpenCL support to run the algorithm in both CPU and GPU. These algorithms deal with image
data stored in 2D matrices. All the tests were done on every frame of a one-minute long video,
and the time taken to process all the frames were recorded. The test results show that turning
the debug option makes the algorithms run faster. Furthermore, bilateral filtering shows about
3 minutes of reduction in execution time when using both CPU and GPU, while other algorithms
show minor reductions in execution time except face detection.
