Research Seminar Series: Bayesian multi-object tracking for autonomous system applications
A/Prof. Reza Hoseinnezhad
School of Engineering, RMIT University, Melbourne
In many autonomous systems, machine intelligence involves estimating and tracking of the states of multiple objects from sensor measurements. This “multi-object tracking” problem is especially challenging when, in addition to the states, the number of objects are unknown and randomly vary with time, and the measurements include misses and false alarms. Such problems arise in a host of application areas including aerospace, defense, field robotics, communications, environmental, and biomedical research. With the proliferation of advanced yet inexpensive sensing technologies (e.g. Kinect), efficient solutions for multi-object tracking problems involved in autonomous systems are increasingly in demand. The most widely known class of such solutions are Bayesian multi-object filters that generalize classical paradigms such as Kalman/particle filtering to multi-object systems. The last decade has witnessed exciting developments in Bayesian multi-object filtering theory and practice. The introduction of finite set statistics and random finite set theories to signal processing has led to the development of the Probability Hypothesis Density (PHD) filters, which attracted substantial interest from academia and industry alike. The PHD filters have been used by BP in oil pipeline tracking, DSO Singapore in passive sensing, FGAN in the 2007 NATO ‘Bold Avenger’ defence exercise and Lockheed Martin in the US space fence program. More recently, multi-Bernoulli and labeled multi-Bernoulli filters have proven to be more versatile and accurate in solving multi-object tracking problems. This seminar presents a general introduction to application of random finite set filters in multi-object tracking problems, with a particular focus on the recent multi-Bernoulli solutions for visual multi-target tracking in autonomous systems. This is a fertile area of research and many aspects and applications have not been explored yet. The workshop is generally targeted at postgraduate students and researchers as well as all those who have an interest in stochastic filtering and tracking. The talk is expected to inspire the audience towards fresh ideas to apply the new approaches for developing efficient solutions for their research problems.
Reza Hoseinnezhad received his BSc, MSc and PhD degrees, in Electronic, Control and Electrical Engineering, all from the University of Tehran (Iran), in 1994, 1996 and 2002 respectively. Since 2002, Reza has been working in various academic positions at the University of Tehran, Swinburne University of Technology, The University of Melbourne, and RMIT University. Reza is currently an Associate Professor, Program Manager for Advanced Manufacturing and Mechatronics Engineering (Honours), and a member of the Academic Board at RMIT University. He is currently a Chief Investigator on two ARC Discovery Project and three ARC Linkage Project grants, leading one of them. He has also received funding from many Government and Industry bodies such as DSTO, Australian Meat Processors Corporation (AMPC), and SMEs. His research interests include machine vision for autonomous systems, perception and control systems for autonomous vehicles, stochastic multi-object filtering, multi-target tracking, statistical information fusion, and sensor management. His significant publications and current projects can be viewed in his website.
For more information, please contact Professor Chun Wang.