Domain Specific Optimisation Techniques for Real-time Image/Signal Processing on Heterogeneous FPGA+CPU platform

The increasing pervasion of vision systems in a number of application fields like machine vision, autonomous vehicles and systems, wearable and mobile devices, object detection and tracking, brings with it the need for the development of efficient and low power data and image processing capabilities. The speed and complexity of computation required dictate hardware based solutions to ensure fastest possible execution of algorithms with the lowest power requirement. Typically these types of solutions have been investigated through CPU/GPU development systems and requiring a complex and custom process for compiling to silicon for an efficient implementation of the end system.

The availability of more advanced hardware development platforms combining FPGA, CPUs and GPUs, with various interface and such as the Xilinx Zynq Ultrascale+ MPSoC lead to an interesting and timely point in this development flow. Such systems can bring together some of the flexibility benefits of high level computation through CPUs while retaining the efficiency and speed of direct hardware implementations. What needs to be understood and researched is the most efficient way to develop and implement systems in this heterogeneous computing platform. Research will focus on the development of domain specific optimization techniques for image processing accelerators and computing methods to support the image processing and analysis requirements of developing markets allied to the industrial partner’s activities.

Jointly sponsored by STMicroelectronics Imaging Division and the University of Stirling, the project will be focussed on real industrial and product development requirements and will require the implementation of key algorithms with stringent power, memory and speed efficiency targets. These developments will, in the first instance, feed and support the technology development process for STMicroelectronics Imaging Division, but will also develop into the more complex aspects of object detection, classification, tracking and segmentation – to feed into product development capabilities.

This PhD studentship will be attractive to students with a strong interest in digital systems design coupled with image processing and computational efficiency. Working with an industrial partner you will be supported with state of the art technology platforms and exposed to the product and technology development processes involved in bringing to the marketplace modern vision systems for markets serviced by STMicroelectronics.

For informal inquiries, contact Dr. Deepayan Bhowmik, University of Stirling ([Email Address Removed]) or Dr. Robert Nicol, STMicroelectronics ([Email Address Removed]). The candidate will be primarily based in the Division of Computing Science & Mathematics, University Stirling with frequent visit to STMicroelectronics, Edinburgh.