In recent years, there has been a large and ever increasing growth interest in computer vision applications for embedded portable IoT devices, while at the same time satisfying high performance, low computing cost and small storage requirements. The use of deep learning approaches increases the performance of such systems at very high levels having also demanding memory and power requirements.
The main purpose of the project is to design and implement an heterogeneous platform consisting of multiple Low Power Graphic processors with deep learning algorithms acceleration capabilities. A software SDK for vision problems using optimized deep sparse coding techniques especially designed for the aforementioned platform will be provided. The significant benefits of the proposed deep eviot system extends to a very wide range of applications from industrial and surgical robotics to autonomous vehicles, smart security cameras and military applications.