Centre for Intelligent Systems Research

CISR Research Seminar Series - 2014

  CISR presentation
blue star Professional development
yellow star Keynote lecture
purple star External presentation
red star No presentation

Seminars will be held at 12pm in the CISR Breakout Area (except where otherwise indicated)

Date Presenter Presentation/topic
Monday 14th April Sahar Araghi Intelligent Traffic Signal Timing Control Using Machine Learning Methods 

 Abstract

The increasing amount of traffic in cities has a significant effect on the road traffic congestion and therefore the time it takes to reach a certain destination, the amount of air pollution and related disease. Extending roads and increasing their capacity is not a sufficient solution, as there will be always an end point, like bottlenecks or intersections. Although bottlenecks cannot be prevented, there is a lot of room for the way intersections are controlled. A common way to control the intersections is using traffic signal light and adjusting the time of each traffic phase. In my research, machine learning methods are applied to control signal timing.

Monday 7th April

5:50pm
(na1.418)
Prof. Bill Moran External presentation  The Ubiquitous Sensor 

 Biography

Professor Bill Moran, from the Department of Electrical and Electronic Engineering, has been appointed as Director of the Defence Science Institute (DSI), a joint venture between the University of Melbourne and the Defence Science and Technology Organisation (DSTO).

Professor Moran is an expert in radar technology, coding and information theory, waveform adaptive sensing, information geometry and compressive sensing, high resolution radar for environmental monitoring, scalable robust video surveillance over constrained networks, mathematics of distributed radar, radar on a chip (ROACH), detection and tracking of targets using distributed antenna, sonar simulation modelling, rapid prototyping, and sensor networks.

Professor Moran is a Fellow of the Australian Academy of Science (FAA), a Member of the Institute of Electrical and Electronic Engineers, and a Member of the London, Australian and American Mathematical Societies.

 Abstract

Sensors are becoming an increasingly important part of our society. Cameras, radars, IR sensors, microphones, are everywhere. If correctly used in disaster management contexts such as bushfires they would be able to assist in deployment of first responders and evacuation of residents.

The aim of this talk is to discuss the theory of sensing, mostly in fairly general terms but with examples taken from disaster management and defence. One aspect of sensing that is being considered in the research community is adaptivity. Sensors can change in many ways: cameras can move, change focal length, change aperture. Radars can change their illumination pattern.

How can we better use this adaptive aspect of sensors to extract the most information from a scene? What is information anyway? And how much does it cost to collect? How can we automate the adaptivity of sensing to optimize the results.

Monday 7th April

5:10pm
(na1.418)
Prof. Laszlo T. Koczy External presentation  Fuzzy signatures 

 Biography

Professor and President of the University Research Council, Szechenyi Istvan University (SZE, Gyor) and Budapest University of Technology and Economics (BME) Hungary

Laszlo Koczy received the M.Sc., M.Phil. and Ph.D. degrees from the Technical University of Budapest (BME) in 1975, 1976 and 1977, respectively; and the (postdoctoral) D.Sc. degree from the Hungarian Academy of Science, all in Electrical/Control Engineering. He spent most of his career at BME until 2001 and from 2002 at SZE. However, he has been a visiting professor at various universities abroad, namely in Australia (ANU, Murdoch and UNSW), Japan (TIT), Korea (POSTECH), Austria (J. Kepler U.), Italy (U. of Trento) and Brazil, China, Finland and Poland for summer schools. He was one of the LIFE Endowed Fuzzy Theory Chair Professors at Tokyo Institute of Technology and advisor to the Laboratory for International Fuzzy Engineering Research in Yokohama. His focus of research interest is fuzzy systems and Computational Intelligence topics (evolutionary algorithms, neural networks), as well as applications. He has published over 370 refereed papers and several textbooks on the subject. He introduced the concept of rule interpolation in sparse fuzzy models, and applied it successfully to the control of an automatic guided vehicle; further hierarchical interpolative fuzzy systems and fuzzy Hough transform. This latter provided the key technology in the winning vehicle in the 2007 Hungarian Mars Rover Competition. His research interests include applications of CI for telecommunication, transportation, vehicles and mobile robots, control, information retrieval, etc.

Among others he had been an Associate Editor of IEEE TFS and he is an Associate Editor of Fuzzy Sets and Systems, Int. J. of Fuzzy Systems, J. of Advanced Computational Intelligence, Mathware and Soft Computing, etc.

He was the General Chair of FUZZ-IEEE 2004 in Budapest, and a number of other conferences, co-chair, PC member, etc. at many other scientific events. He served in the International Fuzzy Systems Association as President, and is now Administrative Committee member of IEEE Computational Intelligence Society.

At SZE he serves his second term as Dean of Engineering, he chairs the Ph.D. School Council and is one of the sponsors of the Szechenyi Alternative Fuel Engine Vehicles Competition, the National Conference of Mechanical Engineering Students, etc.

 Abstract

Fuzzy signatures (FS) are complex structured uncertain descriptors which are suitable for manipulations even when their respective actual structures are not entirely identical. This presentation will give an introduction to the definitions and basic operations in connection with FS.

In many engineering problems there is a series of features which may be grouped into subsets with components related closer to each other, even to sub-subsets within these subsets. Such structures may be represented by either a tree graph, or an iteratively nested vector (with sub-vectors as components).

A very special extension of the idea of FS is given by the Fuzzy Situational Maps (FSM) where the sub-trees represent matrices of two or more dimensions with more or less fixed spatial structure. Zoom in and zoom out operations combined with proper fuzzy aggregations help to increase or decrease the detail view of a given part of the area described by the FSM.

A series of possible applications of FSM will be presented such as description of condition of residential buildings, warehouse layouts and scenarios for intelligent collaborating robots.

Monday 7th April

(na1.417)
Tim Hancock Professional Development  Human Research Ethics at Deakin University 
Monday 31st March Hussein Haggag LGT/VOT Tracking and Performance Evaluation of Depth Images 

 Abstract

This presentation presents object tracking in depth, RGB and normal-maps images using LGT tracker. The depth and RGB images are rendered using depth imaging plugins. A series of experiments were held to evaluate the tracker performance in tracking objects in different image sequences. The experiments conducted were from the Visual Object Tracking (VOT) challenge that was arranged in association with ICCV'13. The accuracy was chosen as the evaluation measure, where the the tracker's bounding box was compared against the ground truth bounding box. Results show that tracking object using depth images gives better results and is more accurate than tracking using either the RGB or normal maps images.

Monday 24th March Fuleah Abdul Razzaq Non-Uniform Sparsity in MRI 

 Abstract

Magnetic Resonance Imaging(MRI) is one of the mostly used imaging techniques in hospitals for capturing images of human body for disease diagnosis and analysis. It differentiates very well between different kinds of tissues which makes it very useful for brain and cancer imaging. Ideally, MRI can be used for capturing live video stream which can be used during surgery, for diagnosis and educational purposes. However, there are some limitations in MR imaging. The imaging process is slow and bound to hardware constraints. It is costly in terms of time as well as motion sensitive which makes it hard for patients. My work contributes towards improving MR imaging process in terms of imaging speed and quality. Enhancing software capabilities can overcome hardware limitations to some extent. This is work is based on the software, signal and image processing module of MRI.

This research explores sparsity distribution MR images. Sparsity of any image can be defined as the information content in that image. MR machines capture Fourier signals which are later converted into images. The first part of my work analyses and identifies sparsity distribution of MR images. Different kinds of Images are used for analysis to understand sparsity distribution in more generic ways rather than making it application specific. Moreover, sparsity is also analysed in different domains other than image and Fourier. The experiments were further extended to localising the sparsity with sub-region of images thus, getting a better understanding of non-uniform nature of MR image sparsity.

The second part presents a novel method to use localise sparsity for MR image de-noising. MR images are corrupted by random Gaussian or Rician Noise. The proposed technique use a simple method to remove this noise based on rules and understanding of localised sparsitywhich was developed earlier. This method analyses and preserves energy contents of imageafter dividing it into a multiple local sections. The simple idea behind this technique is to maximise energy while minimizing the number of non-zero coefficients. Thus, discarding as much noise data as possible and keeping only few carefully chosen coefficients for improved Signal to Noise Ratio (SNR).

The third part uses local sparsity and combines it with Compressive Sensing to achieve Rapid Imaging. The modified proposed approach to Compressive Sensing is named as Locally Sparsified Compressive Sensing. It uses multiple local sparsity constraints and L1 minimisation to reconstruct image from under-sampled data. Measuring fewer samples and reconstructing image from under-sampled data means reducing the image acquisition time and delays caused by MRI hardware. Moreover, a structured framework is presented to define shape, size n number of regions to use Compressive Sensing with local sparsity constraints. Different kinds of MR images were used for experiments and results were compared to simple Compressive Sensing. In comparison to simple Compressive Sensing, this method resulted in reducing sample set up to 30%.

In last part, Locally Sparsified Compressive Sensing was extended for two further applications. Firstly, to improve image quality and decreasing noise occurred due to under-sampled data measurements in simple Compressive Sensing. The basic idea was to use Locally Sparsified Compressive Sensing and exploiting the freedom of using multiple sparsity constraints and sampling levels within an image to improve image quality and reduce noise. Secondly, this developed framework is extended for Dynamic MRI which deals with multiple images captured closely over time to capture some change and motion.

Monday 17th March

2:00pm
CISR Meeting Room
A/Prof. Chee Peng Lim Professional Development  The Craft of Scientific Writing - 3 

 Abstract

Paper writing sessions following the famous Michael Alley's course of "The Craft of Scientific Writing".

Monday 17th March Anwar Hosen Aggregation of PI-based Forecast to Enhance Prediction Accuracy 

 Abstract

In contrast to point forecast, prediction interval-based neural network offers itself as an effective tool to quantify the uncertainty and disturbances that associated with process data. However, single best neural network (NN) does not always guarantee to predict better quality of forecast for different data sets or a whole range of data set. Literature reported that ensemble of NNs using forecast combination produces stable and consistence forecast than single best NN. In this work, a NNs ensemble procedure is introduced to construct better quality of PIs. Weighted averaging forecasts combination mechanism is employed to combine the PI-based forecast. As the key contribution of this paper, a new PI-based cost function is proposed to optimize the individual weights for NN in combination process. An optimization algorithm, named simulated annealing (SA) is used to minimize the PI-based cost function. Finally, the proposed method is examined in two different case studies and compared the results with the individual best NNs and available simple averaging PIs aggregating method. Simulation results demonstrated that the proposed method improved the quality of PIs than individual best NNs and simple averaging ensemble method.

Monday 10th March Mohammed Hossny Video and Image Fusion 

 Abstract

It is not uncommon in many image acquisition applications to resolve a tradeoff between obtaining a high resolution image and a burst of low resolution live images. In this presentation, we introduces a novel framework for imposing the spatial derivative change of low resolution burst sequence to a single high resolution image. The result is a sequence of high resolution warped images. Many application domains such as remote sensing, low radiation live radiology and battlefield automation will benefit from this novel fusion framework.

Monday 3rd March

2:00pm
CISR Meeting Room
A/Prof. Chee Peng Lim Professional Development  The Craft of Scientific Writing - 2 

 Abstract

Paper writing sessions following the famous Michael Alley's course of "The Craft of Scientific Writing".

Monday 24th February Luke Nyhof Non-linear weighted multi-point reference for Adaptive EEG filters: An Application to dynamic motion 

 Abstract

In this presentation, a new method of determining an optimal reference weighting in a two dimensional plane is demonstrated. The focus is on the time-frequency contamination of surface EEG (sEEG) signals by Electromyographic (EMG) noise sources. In particular I propose a new method of minimising the level of signal corruption due to muscle noise based on a cross-correlated multi-point weighted reference system which is then applied to several adaptive filters, focusing on the Weiner Filter and the Least Means Squares adaptive filter. This method has been applied to both simulated EEG and real EEG recorded from healthy subjects. Results show that the proposed method is able to increase the signal-to-noise ratio for EEG signals which are contaminated with muscle noise artefacts; furthermore an application to real biosignal acquisitions recorded during physical movement to a level where analysis has been previously prohibitive.

Monday 17th February

2:00pm
CISR Meeting Room
A/Prof. Chee Peng Lim Professional Development  The Craft of Scientific Writing - 1 

 Abstract

Paper writing sessions following the famous Michael Alley's course of "The Craft of Scientific Writing".

Monday 17th February Houshyar Asadi Human Perception-based Washout Filtering 

 Abstract

Driving simulators are very useful research tools for the governmental institution and research laboratories which are studying in different fields of vehicular and transport design to increase road safety. The aim of this study is to propose the best motion cueing algorithm that can accurately transform vehicle accelerations and angular velocities into simulator platform motions at high fidelity, within the simulator's physical limitations. This is to present the driver with a realistic virtual driving experience and less human sensation error. This presentation will review the various washout filter algorithm architectures, along with the suitable vestibular system models. The review has highlighted the drawbacks and gaps within the different kinds of washout algorithms and vestibular models. Finally, the proposed methodology utilized for the development of an improved optimal motion cueing algorithm is presented.

Monday 10th February Husaini Aza Mohd Adam Colour Identification Based On Haptic Vibrational Frequencies 

 Abstract

The human's visual sensory modality is capable of receiving a large amount of visual information. In today's world, an increasing amount of information is presented visually using digital screen displays. The ability to adequately perceive such visual information has a significant impact on day to day life. An example of such information is 2D visual art, where without adequate vision the information cannot be perceived and the art appreciated. Sensory substitution is one solution to representing visual information to the visually impaired. This paper introduces a haptic system which has been developed to represent colours through haptic vibrations. A new method for mapping colours to vibrations is proposed and evaluated. Vibration representing colour is generated using the Novint Falcon haptic device enabling users to identify colours within a 2D image. A frequency range of 20 Hz to 290Hz is utilised and users are able to differentiate thirteen distinct frequencies corresponding to thirteen colours. The results also show that participants are more successful in differentiating colours towards either end of frequency range than they are in the mid-range which aligns well with observations by other researchers about the frequency response of the human's tactile sensory modality.

Monday 3rd February Sherif Haggag Cepstrum Based Unsupervised Spike Classification 

 Abstract

In this research, we study the effect of feature selection in the spike detection and sorting accuracy. We introduce a new feature representation for neural spikes from multichannel recordings. The features selection plays a significant role in analysing the response of brain neurones. The more precise selection of features leads to a more accurate spike sorting, which can group spikes more precisely into clusters based on the similarity of spikes. Proper spike sorting will enable the association between spikes and neurones. Different with other threshold-based methods, the cepstrum of spike signals is employed in our method to select the candidates of spike features. Simulation results demonstrate that the proposed method not only achieve more accurate clustering results but also reduce computational burden, which implies that it can be applied into real-time spike analysis.

 

Deakin University acknowledges the traditional land owners of present campus sites.

10th April 2014