OpenCV was designed to be cross-platform. So, the library was written in C and this makes OpenCV portable to almost any commercial system, from PowerPC Macs to robotic dogs. Since version 2.0, OpenCV includes its traditional C interface as well as the new C++ one. For the most part, new OpenCV algorithms are now developed in C++. Also wrappers for languages such as Python and Java have been developed to encourage adoption by a wider audience. OpenCV runs on both desktop (Windows, Linux, Android, MacOS, FreeBSD, OpenBSD) and mobile (Android, Maemo, iOS).
Android
Since 2010 OpenCV was ported to the Android environment, it allows to use the full power of the library in mobile applications development.
ARM
Currently, most embedded devices use CPUs based on ARM architecture, including Cortex-A and Cortex-M series. Deep Learning algorithms are usually trained on x86/x64-based servers with powerful Nvidia GPUs. But then the inference needs to be performed on low-power ARM chips.
CUDA
In 2010 a new module that provides GPU acceleration was added to OpenCV. The ‘gpu’ module covers a significant part of the library’s functionality and is still in active development. It is implemented using CUDA and therefore benefits from the CUDA ecosystem, including libraries such as NPP (NVIDIA Performance Primitives). With the addition of CUDA acceleration to OpenCV, developers can run more accurate and sophisticated OpenCV algorithms in real-time on higher-resolution images while consuming less power.
iOS
In 2012 OpenCV development team actively worked on adding extended support for iOS. Full integration is available since version 2.4.2 (2012).
OpenCL
In 2011 a new module providing OpenCL™ accelerations of OpenCV algorithms was added to the library. This enabled OpenCV-based code taking advantage of heterogeneous hardware, in particular utilize potential of discrete and integrated GPUs. Since version 2.4.6 (2013) the official OpenCV WinMegaPack includes the ocl module.
In the 2.4 branch OpenCL-accelerated versions of functions and classes were located in a separate ocl module and in a separate namespace (cv::ocl), and often had different names (e.g. cv::resize() vs cv::ocl::resize() and cv::CascadeClassifier vs cv::ocl::OclCascadeClassifier) that required a separate code branch in user application code. Since OpenCV 3.0 (master branch as of 2013) the OpenCL accelerated branches transparently added to the original API functions and are used automatically when possible/sensible.
