Graphical Models for Sentient Buildings
2 year initial position with opportunities for extension
Australian Centre for Visual Technologies
University of Adelaide, Australia
(with industry partner Snap Network Surveillance)
The Australian Centre for Visual Technologies (ACVT) at the University of Adelaide is one of the largest computer vision and image analysis research groups in Australia. We work on a mixture of fundamental and commercially oriented research projects in computer vision and machine learning, and we have a strong track record of publication in the top venues.
We have recently secured ARC Linkage Project funding to develop novel probabilistic graphical model (PGM) techniques for finding relationships between sensors (and detections made by those sensors) in smart buildings. Visual sensors are of particular interest due to their generality. The project builds on existing scalable technology for finding overlapping sensor fields and thus focuses on the non-overlapping sensors case (again with particular reference to video) to discover an enhanced activity topology for the sensors.
We will develop techniques for learning this enhanced topology on the basis of visual recognition and matching methods that are designed to operate on different views of a person at slightly offset times. By accumulating many candidate matches over time, we can estimate conditional activity distributions over incoming and outgoing links. For example, the path by which a person exits a camera, and the next camera they appear in, may depend on their observed entry point as well as elapsed time. We intend to model such relationships using a novel technique for learning the structure of a probabilistic graphical model based on the observed data.
The project is run by Dr Anthony Dick, Dr Javen Shi and Dr. Henry Detmold (CTO at the industry partner Snap Network Surveillance). It will be conducted in collaboration with Snap Network Surveillance, which is a startup company founded by ACVT in 2009, and which is commercialising ACVT research in large scale video surveillance. The resulting system will be used in future Snap products, and will therefore be a high profile commercial application of vision research.
Within the project there will be considerable room for original research which we expect will lead to publications in computer vision and machine learning conferences and journals. One or more patents will be filed on the basis of the research. There will also be opportunities for joint supervision of postgraduate and honours level students, collaboration with ACVT, with visitors and with both international and industry partners.
The successful applicant must have:
– a PhD in a relevant discipline
– a background in computer vision, and ideally also machine learning, preferably including Online Learning and Probabilistic Graphical Models
– experience in C++ and either Python or Matlab, including the construction of large software systems
– a good publication record commensurate with experience
– fluency in written and spoken English
– the ability to work individually and as a member of a broader team, including with industry partners
Industrial research experience including patent authorship would be an advantage.
Salary: University Level B (AU86,147 – 95,838 depending on qualifications), plus benefits. Some funding for relocation may be available. The ACVT has a generous conference attendance funding scheme for papers accepted to the major conferences.
The closing date for applications is 11:59 PM 16th January, 2015, Australian Central Standard Time (UTC+9.30). Please be sure to include in your application a full CV (including a publication list and statement of research expertise and interests) and the names of two referees.
Please email any queries and/or applications to: Mr Ian Will, Ian.Will@adelaide.edu.au, quoting reference number A02439.
