Computer Vision is growing quickly with many new killer apps like augmented reality, intelligent transportation, 3-D scanners, and no doubt many more advanced interactive systems of the future ... so does it make more sense to define a new programming language for CV like Halide (http://halide-lang.org/) does, to use APIs (as has been done now for some time with OpenCV) or to define hardware co-processors? Or perhaps some mix of the 3 as has evolved for graphics (GL, OpenGL, GPUs, RenderMan, etc.). I'm just curious what researchers favor and see as advantage and disadvantages of each.
Most "new languages" are just scripts that end up being interpreted or translated into source code for C++ like mr Rakhshan had said above.
If you need to do testing and speed is not really too important, but the result is, you will most probably use MatLab or similar software for advanced mathematics and data manipulation and analysis, but when it comes down to production time, you will have to rewrite all your functions and scripts in C/C++ with an adequate library.
My personal preference is OpenCV with C++.
In my opinion, since the final goal of a Computer Vision application is to be integrated or developed into a final "commercial" system, image co-processors are the way to go.
Once an algorithm is implemented and working in (let's sat) C code, there are still a lot of challenges to make it perform nicely in a real system (let's say a mobile device).
An image co-processor approach where the needed functions are nicely packed in a compact implementation (HW+SW) that as fast as it can be in the platform you decide, will simplify the final task of "implementing" the algorithm in a real system.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages