Summer Internship 2010/2011 Project List

Context-Aware Saliency for Scene Analysis
Supervisor: Mehrtash Harandi
Description: The classical way to analyse a scene is to scan the image exhaustively, looking for familiar objects at different locations and sizes. This traditional view can be considerably improved if an attention mechanism detects salient parts of an image/video and directs the analysis process to investigate the more interesting parts.
In practise finding the salient parts of an image has various applications. To name a few one may consider object recognition, abnormal behaviour detection and video abstraction. Amazingly, the notion of saliency has been supported by physiological and neuroscience findings. For example the distribution of photoreceptors on the human retina is highly non-uniform by which only a small region of visual angle around the centre of gaze is captured at high resolution.
The aim of this study is in-depth analysis of state-of-the-art approaches, followed by devising a novel saliency detection method for visual surveillance applications. Applicants should be familiar with the basics of image processing and have a strong mathematical background. Interested applicants are encouraged to check the following references.
[1] S. Goferman, et al., "Context-aware saliency detection," in Computer Vision and Pattern Recognition (CVPR), 2010 IEEE Conference on, 2010, pp. 2376-2383.
[2] L. Itti, et al., "A Model of Saliency-Based Visual Attention for Rapid Scene Analysis," IEEE Trans. Pattern Anal. Mach. Intell., vol. 20, pp. 1254-1259, 1998.

The application of face recognition technology to mobile mixed reality games
Supervisors: Lachlan James, Ricky Robinson & Sandra Mau
Description: NICTA QRL is developing face recognition and associated algorithms for various applications within the broad area of Safety & Security. The goal of this project is to employ these algorithms in a different context: mixed reality games.
A mixed (or alternate) reality game (MRG) combines different media to immerse the player in a narrative, which may evolve in response to the playerʼs actions. Often a MRG will include a mobile element, and draw upon the particular character of the environment in which it is played. MRGs are fast gaining in popularity, having emerged from the realms of research and academia, and having been adopted by mainstream gaming studios.
The major challenge of this project is to incorporate face recognition technology into a mobile mixed reality game based around the Apple iPhone platform. The game will involve some variation on the idea that players are paired with one another, and must attempt to capture a sufficiently clear image of the other playerʼs face using the camera in their iPhone, without letting the other player accomplish the same task. This may occur over a period of days or weeks. Captured images are then submitted to the game server for matching. The game is won when the game server manages to match the captured image with the other playerʼs profile image. Specific challenges include the creation of a mechanism to validate a playerʼs mug shot (profile image) upon registration and finding a means to prevent forgery of images submitted for matching against a playerʼs profile image (by, for example, searching for an existing image of the player on the web, and submitting it). This may involve some form of calibration with other data sources (e.g., geo-location).
Ideally, the successful candidate will have mobile phone development experience and knowledge of RESTful web application design and development. Knowledge of basic statistical methods for data fusion may prove to be an advantage.
Desired qualifications: Experience with iPhone development

Rethinking patient charts for the iPhone era
Supervisors: Lachlan James & Ricky Robinson
Description: In conjunction with the Queensland Health Skills Development Centre, NICTA QRL is exploring ways in which contemporary mobile and personal devices can be coupled with fixed devices (flat screens, terminals and so on) to provide more efficient and reliable in situ interaction with patient health records. This project will investigate and report on the complexities of entering information into, and interpreting data from, patient charts.
Patient charts, generally located at the foot of the patientʼs hospital bed, serve an important role in the treatment of the patient. They record the prescribed medications, vital signs (blood pressure, heart rate and so on), and arbitrary notes made by doctors and nurses. As such, they are a means of information exchange between the hospital staff responsible for treating the particular patient.
Currently, patient charts are static. They provide no immediate feedback, and must be transcribed if a digital record is required. An opportunity exists for simplifying the data entry and output problem. However, to perform a conversion from paper-based patient charts to digital ones a number of obstacles must first be navigated. A range of problems in human- computer interaction present themselves in this environment, and these are compounded by very specific hygiene and other requirements. Though many studies (ethnographic, simulations, and so on) have been performed in the past, it is not clear what an appropriate combination of digital devices and software for this space would be.
The successful candidate will draw upon the existing literature and conduct a short study of the hospital environment to undertake rapid prototyping of a digital patient chart. Access to the Queensland Health Skills Development Centre at the Royal Brisbane Hospital will be organised for the purposes of observation and testing.
Desired qualifications: Experience with developing for the iOS platform (iPhone, iPad) or other similar technologies. User studies.

Exploring and critical review of vision-based people detection algorithms and techniques
Supervisor: Dr. Farhad Dadgostar
Description: The successful applicant will conduct a review of literature, commercial and open source software for people detection from video. This project will require implementation of some of the existing people detection algorithms, and evaluating their performance on the i-LIDS surveillance video dataset. The outcome of this project will be a comparative report on the state of the art techniques for video-based people detection. A good contribution to this subject will earn the coathorship of a conference paper.
Desired qualifications: This position requires good C++ programming skills, and familiarity with fundamentals of computer vision and image processing.

Exploring and review of visual indicators for better monitoring and analysis of multiple surveillance video feeds
Supervisor: Dr. Farhad Dadgostar
Description: In the context of vision-based automated surveillance there has been little work on time-cumulative information which can be extracted from single or multiple video feeds. For instance the object trajectory map or motion heat map can summarize information of hours of video footage to a single image, making it more sensible for non expert observers. In practice these information can be a rich resource from a managerial point of view. The main goal of this project is exploring methods of video summarization and implementing a few of them to demonstrate the proof of concept. The successful applicant will work with a small software engineering team to integrate his/her proof of concept with a lab-based CCTV monitoring station.
Desired qualifications: This position requires good C++ programming skills, familiarity with fundamentals of image processing and computer vision, basic knowledge of computer graphics and familiarity with OpenGL and Qt library.

OpenCV Face Detection on GPU/OpenCL
Supervisor: Dr. Farhad Dadgostar
Description: Face detection on video streams, is an important yet CPU intensive component for face recognition systems. The OpenCV library provides an open source face detector which arguably is the most popular face detector in the research community. The underlying face detection algorithm, however supports multi-core hardware through OpenMP, it technically can be highly parallelized to operate much faster than its current implementation. The goal of this project is reengineering the face detection algorithm using OpenCL. This project would be an excellent opportunity for someone who is seeking experience in CUDA/OpenCL programming and likes working with new technologies. This implementation will be integrated with our distributed video analytics framework. Hence, the successful applicant will collaborate with our engineering team to address the real-world requirements while implementing the solution.
Desired qualifications: This position requires good C++ programming skills, some background in software implementation for GPU – preferably CUDA and OpenCL, basic knowledge of image processing and computer vision.

MAC Layer Detection Mechanism for Node Selfishness in Wireless Mesh Networks
Supervisora: Dr. Marius Portmann & Dr. Vallipuram Muthukkumarasamy
Description: We intend to develop a system that detects misbehaving nodes (selfish or malicious) in the MAC layer of the IEEE 802.11 protocol in the WMN architecture overcoming the shortcomings of the existing mechanisms. We address the MAC layer misbehaviour such as backoff timer manipulation targeting at rectifying the limitations such as false positives associated with interference and hidden terminal problem that exists in the current work. In this sense we develop a wireless sniffer that uses statistical analysis of the captured frames from Mesh Routers (MR) and identify cheaters. We test our scheme based on a conducted test bed which is a more realistic model of the real world environment compared to software simulation. We hypothesise false positives can be reduced if nodes could have an understanding of their neighbours to identify legitimate neighbours that cause collisions by correctly decreasing their backoff timer due to hidden terminal problem. Embedding the Neighbourhood Discovery Protocol (NHDP) by Clausen et al. is a possible remedial action.

Brisbane City Sensor Audit
Supervisors: Markus Rittenbruch & Marcus Foth
Description: In the same way that web development frameworks such as Ruby on Rails, Django and Spring have greatly simplified the development of web applications, the Urban Informatics research area at NICTA is looking into application frameworks for urban ubiquitous computing. These frameworks will make it simpler to build applications that gather data from sensors and other data sources throughout a city, and enable even non-expert programmers to create “mash-ups” of this information.
This summer scholarship project will require the successful applicant to undertake an audit of the data sources/sensors around Brisbane city. Which of these are publicly accessible? Which are private? What technologies are required to access these data sources? What do citizens think about these data sources? Are they aware of them? How do they use them? Are there privacy issues? In addition, the student may choose to undertake a study of current public transport scheduling/time-tabling technologies deployed in the city, and to assess its reliability and usefulness. The study will also seek input from commuters on the kinds of applications they would find useful. The findings from this study will be used to drive the research and development of new tools for commuters. This project will provide the student with valuable experience in designing and conducting user studies, a required skill for those intending to continue studies in pervasive/ubiquitous computing and related fields.
Desired qualifications: Experience with field and user studies

Sensing at the Edge
Supervisors: Markus Rittenbruch & Marcus Foth
Description: The Edge is the new digital culture centre of the State Library of Queensland. The Edge aims to foster the creativity of young people through a broad range of activates and workshops as well as the provision of resources such as a recording studio and a physical computing lab. NICTA has a special relationship with the Edge and runs workshops and projects on how to support novel types of interactions through sensors and microcontrollers, such as the Arduino platform.
The aim of this project is to build a sensor and visualisation platform that collects data about the use and environmental status of the building and it’s inhabitants. This includes the physical status like a visitor counter, an indicator which workspaces are free / busy as well as the “social” status of the building, such as which groups are meeting, whether there are is live music, etc. Based on the captured sensor data the student will design visualisations that display and merge the collected data to effectively convey a particular view, such as the environmental or social status. Visualisations can be screen-based, use actuators, or use interactive means such as our multi-touch table. Data will be distributed using a brokerage platform like Pachube and be made available through the OSC protocol to allow for integration with other existing visualisations at the Edge. The project ties in with our research into application frameworks for urban ubiquitous computing. This project will provide the student with valuable experience in building applied sensor technology and learning how to design visualisations.
We are aiming for the successful applicant to be hosted at the Edge for the duration of the scholarship.

Multi-hop Video Streaming via Google Android Phones
Supervisor: Marius Portmann Description: The goal of the project is to implement and evaluate an ad-hoc (peer-to-peer) video streaming application over multiple Wifi enabled Google Android phones. The video is captured on one device and sent via multiple other phones, which act as relays or routers, to a final destination, where the video is displayed.
NICTA has a basic implementation of this video streaming application, but what is missing is the integration with an Ad-hoc routing protocol, to achieve flexible multi-hop operation.
Desired qualifications:The project requires a keen interest in application development for mobile devices (Android platform), some networking background, and willingness to do experimental work (e.g. performance measurements).

3D modelling of insects using shape-from-shading for biosecurity
Supervisors: Dr. Paul Zhang, A/Prof. Yongsheng Gao
Description: There is an urgent need to develop technologies that can automatically screen dangerous pests, insects and objects threatening Australia's food and environmental biosecurity. NICTA's Biosecurity Project is developing automatic and interactive devices that can effectively assist users to identify and interpret different insect species in situ via mobile access. This technology will create a revolution in pest control, plant health and food quality assurance. NICTA's development of in situ identification and screening of red imported fire ants is funded by the Department of Agriculture, Fisheries and Forestry (DAFF) and Biosecurity Queensland Control Centre (BQCC).
This internship project aims to use shape-from-shading technologies to develop a 3D insect modelling system, which facillitates the insect species identification and descriptions. The 3D insect model will be built from a single 2D image and it will enable the system to manipulate the insect model in 3D space which gives the user and system more robust and informative descriptions of the species for identification. Successful applicants will be responsible to implement one or more shape-from-shading techniques and to test/improve these algorithms on standard images (vase, Mozart, penny, Lena, apple, etc.) and our own insect images.
Desired qualifications: C AND Matlab programming skills are required. Good understanding of 3D geometry and basic understanding of geometrical optics are desired.

Hyper-Heuristic Parallel Framework for Small Protein Optimisation
Supervisors: Duc Nghia Pham, Wayne Pullan, Bela Stantic
Description: Optimisation of proteins is a computationally intensive process. This project will be focussed on developing a C++ framework which will allow any number of computer processors to efficiently participate in, firstly, the conformational energy calculation for a protein and, secondly, applying state-of-the-art machine learning and automated reasoning techniques to identify structural changes which will lead to a lower energy state of the protein.

Language-agnostic process model editor
Supervisor: Marcello La Rosa
Description: Both empirical and anecdotal evidence indicate that enterprises in various industries tend to collect thousands of business process models over time. In that sense, it may not come as a surprise that many organizations find it difficult to keep track of such large amounts of complex process models – an issue that is referred to as “model management”. The problem is amplified by the use of different modeling notations such as EPCs, BPMN, UML Ads, within and across organizations. This situation especially results from organizational mergers or acquisitions.
As part of a collaboration involving QUT and five European Universities, we are developing an Advanced Process Model Repository named APROMORE. APROMORE aims to provide an open and extensible platform to store and disclose business process models of a variety of types and languages, equipped with features specifically designed to deal with large process model collections. For its model editing capabilities, APROMORE relies on the open-source model editor Oryx. However Oryx, like any other process model editor, is tightly-coupled to a (few) specific modeling notation(s). This hinders the development of language-independent plug-ins to facilitate the management of large process model collections (e.g. a plug-in to merge two a set of processes modeled in different languages).
In order to achieve that, the objective of this project is to design, develop and test a component for the Oryx editor that can interface with APROMORE via the use of a canonical process format. This format will be language-independent and will only capture a process model’s structure, thus being agnostic to presentation aspects related to process models. The realization of this component will allow Oryx to be easily customizable to support different modeling notations (e.g. BPMN, EPCs, UML ADs, YAWL) and to transfer the result of a modeling activity from one notation to another. Moreover, it will facilitate the development of APROMORE plug-ins to apply different presentation paradigms on top of a process model structure.
Desired qualifications: GPA above 6, knowledge of Business Process Modeling, Java and Java Script.

Defeasible Logic Theory Editor
Supervisors: Brian Lam & Guido Governatori
Description: Writing rules in RuleML or other standard interchange format can often be a cumbersome task. Thus the need for authoring tools that assist end-user in writing and expressing rules is apparently important.
In this project, based on our ongoing research project SPINdle, student(s) are expected to perform design and develop work of creating a visual editor for editing standard and modal defeasible logic theory with support in different rule interchange format, such as RuleML, OWL and RDF.
Desired qualifications: Familiarity with JAVA, XML and RDF. Knowledge in RuleML and OWL is preferred though not essential.

Title: Business Process Compliance
: BPC-2
>Supervisor: Dr Guido Governatori
Description: Compliance of a business processes is fast growing area of research with an high potential of applications. The market value for compliance is estimated to be over 32bilion US dollars in 2008. Compliance is the relationship between two sets of specifications: "normative" specifications describing what a process should do and business process specifications describing the activities a process is going to perform. The aim of the project is to write API to integrate YAWL a powerful and open source workflow system and FCL a rule engine designed to model normative specifications.
Desired qualifications: Familiarity with Business Process Modelling