A large semiconductor company wanted to provide their customers with a high-level native implementation of the popular OpenCV library to ease the flow from development to deployment. The main porting challenge was how to map reference library code targeting x86 SMP onto their own heterogeneous Advanced Driver Assistance System architecture while still achieving real-time image processing.
By using Prism Technology, CriticalBlue engineers could identify potential performance pitfalls upfront during the library porting phase. As a result, the code analysis report and optimized OpenCV implementation were delivered on time while meeting all customer performance requirements.
This led to CriticalBlue recommending and then being commissioned to implement a memory management layer below OpenCV to abstract away complexity at the hardware layer, allowing both the silicon vendor and their customers to make more efficient use of the platform.
A specific benchmark application to be run in real time was developed as a performance guide, making use of two fundamental computer vision algorithms along with some other simple applications.
The project then involved mapping OpenCV functions and data structures onto an existing platform-specific SDK to access hardware acceleration features, followed by further Prism analysis and optimization of the SDK itself.