Supporting localized OpenVX kernel execution for efficient computer vision application development on STHORM many-core platform

Nowadays Embedded Computer Vision (ECV) is considered a technology enabler for next generation killer apps, and scientific and industrial communities are showing a growing interest in developing applications on high-end embedded systems. Modern many-core accelerators are a promising target for running common ECV algorithms, since their architectural features are particularly suitable in terms of data access patterns and program control flow. In this work we propose a set of software optimization techniques, mainly based on data tiling and local buffering policies, which are specifically targeted to accelerate the execution of OpenVX-based ECV applications by exploiting the memory hierarchy of STHORM many-core accelerator.