Fully funded Research Associate position at Nanyang Technological University (NTU), Singapore

A fully funded research associate position is available for candidates with a keen interest in the broad areas of system identification, and in particular, statistical signal processing, information fusion and Bayesian methods. Suitable candidates would potentially hold a Ph.D degree in engineering or applied math, and are expected to have a strong mathematical background, respective skills, and a passion to conduct a cutting edge research. This research is planned for 3‐years and would be carried out under the guidance of Prof. Avishy Carmi in the school of Mechanical and Aerospace Engineering, NTU. A brief description of the scope is provided below.

Bayesian Perceptual Tracking and Reasoning
In this project we aim to provide new advances in computational methods for reasoning about many objects that evolve in a scene over time. Information about such objects arrives, typically in a realtime data feed, from sensors such as radar, sonar, light detection and ranging (LIDAR) and video cameras. A typical traditional scenario would involve one or more moving ships or aeroplanes being observed by a radar or video sensor, in the presence of false measurements, or ‘clutter’, though we are equally concerned here with ecological tracking data, biological data and video sequence data. Traditional reasoning engines, or ‘trackers’, treat each object in the scene as a single entity which does not interact significantly with other objects in the scene, and perform tracking independently over the various sensors which are available. ‘Fusion’ of track information then occurs, in order to form a single ‘picture’ of the multi‐object scene, and finally any decision‐making steps are carried out based on this picture.

As distinct from conventional methodologies, the proposed project involves integrating “low level” identification and tracking techniques with “high level” perceptual capabilities and decision making procedures. In particular, we seek to learn and model interdependencies between objects in a scene through group interactions, graph structures and causality inference, hence obtaining better tracking performance and learning important higher level concepts such as intentionality and behavioural traits automatically from the scene. This is all carried out through state of the art Bayesian dynamical models and using advanced Bayesian computational engines. This project will generate new methods for automated analysis of complex evolving scenes. We will focus on the joint modelling of the objects in the scene through dynamic interaction models which can capture in generic ways the dependencies between objects within particular groupings in the scene. We will also provide an integrated framework for fusion of multiple mobile cooperative sensor platforms (UAVs), which are often asynchronous and not properly grid‐locked, into a single reasoning engine that produces a single picture of the scene and represents the probabilistic uncertainty about the scene in a user friendly way.

Contact Person: Dr Avishy Carmi (ACarmi@ntu.edu.sg), School of MAE, NTU.