Eindhoven University of Technology has a vacancy for a
PhD student on Scalable Video Analysis on low-power GPU architectures
within the Dutch ZERO program on energy automous systems; in particular within the ARM project within ZERO, on autonomous roadside monitoring. As a member of the Electronic Systems group, department of Electrical Engineering, your working week will be divided over two locations: most days at the company ViNotion B.V. and part of the time at the TU/e university.
The advances in video analytics are going in a rapid pace, especially since the introduction of deep learning techniques. Deep learning is based on a neural network comprising multiple layers of connected neurons that can be trained to classify input signals. In the domain of video analysis, this technique is used to detect and recognize or classify object. The neural network has a huge amount of inherent parallelism that can be exploited by parallel processing units at different levels, but due to recent developments and optimizations applied to deep neural networks, this is not trivial.
ViNotion uses state-of-the-art deep learning video analysis techniques to develop innovative real-time surveillance systems for pedestrian, traffic and vessel monitoring (www.vinotion.nl).
The newest video analysis techniques allow sophisticated traffic management to improve the safety and the efficiency of human observation.
Surveillance systems to control traffic on highways typically contain thousands of cameras to collect data and to control the traffic. To make the video analysis scalable, and to limit the bandwidth that is needed for video streaming, video analysis in or close to the camera is needed; it reduces the visual data to the essential information that is needed for the traffic operation. This implies embedded analytics into the camera or an outdoor cabinet. Moreover, in many use cases, no wired power source is available. Hence, there is a need for low power implementations that can operate on solar energy and/or portable batteries. At the TU/e research is performed to efficiently map deep neural networks to various hardware and processing platforms, including GPUs, FPGAs and ASICs (parse.ele.tue.nl).
The PhD will research the architecture of deep-learning classification algorithms and find possibilities to systematically exploit the parallelism in a scalable fashion. On the hardware side, parallelism at instruction level, at task/process level, but also at many-core and multi-processor level should be considered. On the software side, data partitioning should be investigated besides the functional partitioning and efficient mapping of neurons, layers and the total neural network w.r.t. other video processing functions in the system. For complex problems that cannot be computed in real-time on a single GPU, an approach that scales to multiple GPUs is desired.
Eindhoven University of Technology (TU/e)
Eindhoven University of Technology (TU/e) is a world-leading research university specializing in engineering science & technology. The Department of Electrical Engineering is responsible for research and education in Electrical Engineering. The discipline covers technologies and electrical phenomena involved in computer engineering, information processing, energy transfer and telecommunication. The department strives for societal relevance through an emphasis on the fields of smart sustainable systems, the connected world and care & cure. The TU/e is the world's best-performing research university in terms of research cooperation with industry (#1 since 2009).
The Electronic Systems (ES) group consists of six full professors, two associate professors, eight assistant professors, several postdocs, about 30 PDEng and PhD candidates and support staff. The ES group is world-renowned for its design automation and embedded systems research. It is our ambition to provide a scientific basis for design trajectories of electronic systems, ranging from digital circuits to cyber-physical systems. The trajectories are constructive and lead to high quality, cost-effective systems with predictable properties (functionality, timing, reliability, power dissipation, and cost). Design trajectories for applications that have strict real-time requirements and stringent power constraints are an explicit focus point of the group. Within this area, prof.dr. H. Corporaal and dr.ir. S. Stuijk have developed various novel power efficient computer architectures and their associated compilation trajectories.
The ZERO-ARM project team is designed to combine extensive knowledge in the key fields (video processing, compilers and computer architectures). The project team includes industrial partner ViNotion. As part of this project, the PhD candidate will work closely with these industrial partners and will be allocated several days per week - depending on the need - at the office of the partner, to ensure that the developed architecture fits the needs of ViNotion and to quickly learn about the video analysis algorithms.
ViNotion is an innovative development company for computer vision applications, focusing on software development. The company has 10-15 employees, which most have a master of PhD degree in advanced signal processing. For its R&D, ViNotion works closely together with the Eindhoven University of Technology that is located 3 km from the office.
ViNotion offers products and services for crowd analysis and traffic analysis with intelligent video camera systems. The products and services can not only count but are also able to distinguish pedestrians from bicycles and to measure speed and density of people.
These products are being applied for city marketing, retail footfall, crowd and event management, train stations, mobility reporting, urban planning, traffic control, tolling, etc. As a service, ViNotion provides mobility services and offers crowd and traffic analysis systems.
We offer a fixed-term, 4 year position in a research group with an excellent reputation. Salary and benefits are in accordance with the Collective Labour Agreement for Dutch Universities, including:
If interested, please use 'apply now'-button at the top of this page. You should upload the following:
Candidates will be selected based on graduation mark and proficiency at university including consideration of the reputation of the university, relevant experience and skills, writing skills and publications, work experience as well as performance in relevant modeling exercises and interviews.