This project aims to develop the world-first automated asset auditing technology. Dictionary learning and deep learning algorithms have shown promising results in detecting, tracking and understanding assets (or objects in general) in public places for smart city operations. Using such learning techniques, we intend to create new models for automated (1) identification of city assets, (2) determine the condition of the assets (damaged, broken, or missing), and (3) integrate with city services for decision-making and service delivery.
Significance: Our proposed project will fundamentally change the practice of auditing assets. Our solution will deliver the state-of-the-art dictionary learning and deep learning technologies by integrating multi-modal sensing technologies to create dictionaries of assets in existing cities. The outcomes of the project will provide automated tools for auditing of cities and efficient operations of cities. Our innovation is a novel multi-level solution framework for detecting, recognizing and characterizing cities assets.
The real-time IoT applications have stringent delay requirements when implemented over distributed sensing and communication networks. Their finite-time performance matters much more than asymptotic results in the literature. A good illustrative example is smart traffic control, where sensor-laden vehicles pass through intersections by communicating and remaining at a safe distance from each other, rather than grinding to a halt at traffic lights. Thanks to increased investment in smart city infrastructure and projected penetration of autonomous vehicles, smart intersections are expected to replace traditional traffic lights and become prevalent in the near future. Safe and efficient operation of such systems requires control actions to be taken by each system agent, ie, automated vehicles and road side units, within fraction of a second under local communication and computing limitations. The IoT agents in this scenario clearly must attain global objectives of safety and efficiency in real time.
The overarching aim of the project is to develop new practical algorithms for real-time IoT applications and to investigate novel methods for describing their performance in a finite time. This project aims to investigate fundamental performance guarantees of real-time IoT systems using methodology from networking, optimisation, game theory, and information theory.
Further infomration available at: Smart Transport OptimizationThe Internet of Things (IoT) creates a network infrastructure that enables a wide range of physical objects and environments to be monitored in fine spatial and temporal detail. IoT applications that model and interpret the large volumes of complex data streams, the so-called “Big Data”, generated from sensors, video networks and social activities provides a basis for new business and government applications in areas such as retail chains, transport logistics and public safety. Currently, the key challenge lies in the ability to automate the timely interpretation of this data. Manual inspection of this data is impractical due to large volume of data and time consuming. This project aims to discover the hidden relationships in heterogeneous Big Data by revealing deep structures in data streams – clustering to identify natural groupings of objects in the Big Data where no ground truths are available.
Further infomration available at: Big Data Analytics and AIOne in six people worldwide experience a stroke in their lifetime. 15 million people suffer from stroke and 5.8 million die each year. An estimated 6.7 million deaths were due to stroke in 2015. Computed Tomography (CT) scans are used by clinicians to assess the extent of stroke and provide right treatment to save brain tissue. However, current neuroimaging techniques have variable diagnostics accuracies. The project aims to develop new Deep Learning-based Artificial Intelligence (AI) tools to provide better precision for diagnosing stroke patients. The project would also utilise Magnetic Resonance Imaging (MRI) scans (as a gold standard) to validate the new models.
Several neurological disorders affect the motor system in the brain. This results in deprivation of purposeful movement by the patient and affects the normal interaction with the environment. The specialized nerve cells in the lower and upper motor neurons of the brain control the skeletal muscles. Any neurological problem affecting the motor neurons will result in the manifestation of the problem in the hands and limbs. The symptoms may include tremors, rigidity, spasticity, weakness and inaccurate movement. One of the milestones of modern management of acute stroke is the administration of a thrombolytic (clot-busting medication) to unblock the artery. Continuous monitoring of patients is critical in the management of stroke patients. However, the current clinical observation paradigm is time consuming and subject to inter-observer bias. The project aims to develop wearable devices with Edge Computing and Artificial Intelligence (AI) models to understand and predict patient recovery.
Further infomration available at: Edge AI Wearable DevicesStillbirth, defined as the loss of life after 28 weeks gestation, affects almost 3 million families across the globe and ends over 2000 Australian pregnancies each year. Sadly in Australia, half of all stillbirths occur after 37 weeks gestation with the incidence of stillbirth rising drastically with each week. This is particularly devastating, given the baby could often have been safely delivered, if only fetal distress were identified.
In this project we seek to develop new signal processing and machine learning methods to intelligently monitor fetal wellbeing using wearable devices. This will consist of developing computationally efficient algorithms suitable for operation in an Internet-of-Things (IoT) framework. This work continues over a decade of research our group has undertaken developing signal processing techniques and models to better understand maternal-fetal physiology, and builds upon collaborations with The University of Oxford, Tohoku University and The Mercy Hospital for Women.
Further infomration available at: Fetal Monitoring
Internet of ThingsCurrently there are about 20 billion IoT devices connected to the Internet. By 2020, an estimated 50 billion devices will be connected. Already 40% of the 7 billion population have internet connection. By 2050, 70% of the world's population and over 6 billion people are expected to live in cities and surrounding regions. Managing city, people and resources (water, electricity, air, land, transport, public health) are set to become challenging. Technical challenges include designing new embedded IoT devices, communication protocols, network management, analytics, security and privacy, scalability, sustainability and many other. Our group researches on these challenging IoT issues in collaboration with academia, industry and governments. |
|
Grants: ARC Linkage Projects (LP120100529)- "Creating Smart City through Internet of Things"; ARC LIEF (LE120100129) - "Internet of Things Testbed for Creating a Smart City"; EU FP7 SocioTal - "Creating a Socially Aware Citizen-centric Internet of Things" and H2020 Organicity - "Co-creating Smart Cities of the Future" Collaborators: The University of Sydney, Australia; Swinburne University of technology, Australia; Deakin University, Australia; Nanjing University of Posts and Telecommunications, China; University of Surrey, UK; Institute for Infocomm Research, Singapore; Partners: Defence Science & Technology (DSTO); City of Melbourne; Arup; SenSen Networks, Pty Ltd Outcomes:IoT framework for creating a smart city, QoS solution using IoT, framework for handling IoT issues, such as security, privacy and governance. Publications:
|
Event Analytics in Big DataThe current Internet is evolving into the Internet of Things (IoT), where networked devices have the ability to compute, sense and interact with their surroundings. IoT deployments will generate a vast amount of data, but the value of the data lies in how we can exploit it. Currently there is little domain expertise to automate this Big data analysis, and traditional supervised machine learning techniques suffer from a lack of labelled training data in this context. The aim is to develop scalable distributed and incremental event detection techniques in addition to privacy-preserving data analytics solutions for the IoT. Also enables a new generation of applications in IoT applications such as smart cities. We have developed algorithms to detect anomalies in traffic patterns of car using taxi data. Novel incremental learning algorithms have been designed and developed to this end. The solution can be used to monitor parks, urban canopies, plants and tree species. We have also developed a solution with state-of-the-art interactive visualization for new touch-enables devices such as smartphones and tablets. This provides councils, administrators and policymakers to assess the situation in real-time, enabling to make informed and immediate decisions. |
|
Grants: ARC Linkage Projects (LP120100529)- "Creating Smart City through Internet of Things"; ARC LIEF (LE120100129) - "Internet of Things Testbed for Creating a Smart City"; EU FP7 SocioTal - "Creating a Socially Aware Citizen-centric Internet of Things" and H2020 Organicity - "Co-creating Smart Cities of the Future" Collaborators: University of Surrey, UK; Michigan Technological University, USA; Deakin University, Australia; Partners: Defence Science & Technology (DSTO); City of Melbourne; Arup; SenSen Networks, Pty Ltd Outcomes: Real-time event analytics capable of detecting anomalies from IoT data, sensor drift correction, incrementally learning from streaming data, spatiotemporal estimation of missing data, capability to interact with users for real-time decision making.
|
Energy Monitoring:It is expected that the world energy consumption in residential sector is set to increase by 48% from 2012 to 2040. By 2040, around 13% of world energy consumption would account for homes, buildings and non-transport. We are developing algorithms to monitor energy consumption of the buildings and office spaces to understand the nature of energy consumption. We are also working to develop solutions to intelligently integrate smart grid and residential energy usage. |
|
Collaborators: Deakin University, Australia; Univeristy of Twente, The Netherlands; Delft University of Technology, The Netherlands; Politecnico di Milano, Italy; NICTA Victoria Research Laboratory, Australia; University of Surrey, UK Outcomes: We have developed optimization techniques and machine learning models for energy monitoring Publications:
|
Security and Privacy in IoT:The IoT consists of islands of sensor networks are meant to operate unattended, potentially in harsh and even hostile environments. Security is thus essential to ensuring the intended operation of the IoT. For businesses and consumers, security enables the IoT to be used with confidence. Privacy is another paradigm where data are collected from citizen engagement and participatory sensing. Clearly, here the data can contain sensitive information, such as participants' location, voice and other proprietary audio-visual information. We are investigating privacy-preserving data mining, that is, we are addressing the important topic of preserving users' privacy to allow a third-party data miner mines user-contributed data for useful information. |
|
Grants: ARC LIEF (LE120100129) - "Internet of Things Testbed for Creating a Smart City"; EU FP7 SmartSantander; EU FP7 SocioTal - "Creating a Socially Aware Citizen-centric Internet of Things" and H2020 Organicity - "Co-creating Smart Cities of the Future" Collaborators: University of Twente, The Netherlands; National University of Kaohsiung, Taiwan; Università di Roma, Italy; Northwestern Polytechnical University, China; South China Normal University, China; University of Oxford, UK; Delft University of Technology, The Netherlands; University of South Australia, Australia; The University of New South Wales, Australia; Swinburne University of Technology, Australia; Partners: City of Melbourne; Arup; Agency for Science, Technology and Research, Singapore. Outcomes: Algorithms, learning models and techniques for security and privacy in critical infrastructure and IoT. These include multicast authentication, secure reprogramming, and resilience to smart jamming, cryptographic key management, secure routing, and secure data aggregation Publications:
|
Real-time Video Analytics:The prevalence of camera networks for surveillance, together with the decreasing cost of infrastructure, has produced a significant demand for robust monitoring systems. Current systems offer limited functionality, particularly in their reliance on centralised processing of gathered information. This project addresses end-to-end system challenges of camera networks. Integrating developments across the spatial, spatiotemporal and decision domains, the project incorporates distributed sensor network technology with intelligent formation fusion to deliver unique long-term behaviour analysis in highly crowded environments. We have developed video analytics with capabilities to count people, track, and detect suspicious behaviour, suitable for crowd management, modelling, and urban planning. |
|
Grants: ARC Linkage Projects (LP100200430) - "Design of Adaptive Learning Visual Sensor Networks for Crowd Modelling in High-Density and Occluded Scenarios" Partners: City of Melbourne, Melbourne Cricket Club, Arup, and SenSen Networks Pty Ltd Outcomes: Real-time video analytics researched and developed specifically for crowd monitoring in crowded scenes using low-quality videos. It is able to count people, track crowd movements, detect suspicious behaviour (such as loitering). It uses state-of-the art techniques and cutting-edge, novel analytics, unique to crowd monitoring problems. Publications:
|
Blind-Aid Vision:In Australia, there are around 357,000 people who are blind or have low vision. Traditional white cane solution used by the elderly with vision impairment is inefficient as many are dependent on ambulatory aids, such as wheelchairs and walking frames. Uneven surfaces such as potholes and drop-offs cause non-protruding hazards. Current solutions based on proximity sensing is unsuitable for detecting non-protruding hazards. We have developed a proof-of-concept system to address this need. The system developed uses an optical laser system to identify the non-protruding hazards. It combines laser projection pattern, computer vision, pattern recognition, and machine learning. This was developed using a seed grant from the Ian Potter Foundation. |
|
Grants:
Collaborators: Tata Consultancy Services, India Partners: Vision Australia; Guide Dogs, Australia Outcomes: A portable prototype system for assisting disabled and elderly people that detects potholes and hazards (such as uneven surfaces). It uses optics, computer vision and machine learning techniques to deliver the results Publications:
Media:
|
Biomedical Image Processing:Parkinson’s disease (PD) is a progressive neuro-degenerative disorder increasingly prevalent over the age of 65. In Australia, about 80,000 Australians suffer from PD. Changes in voice are amongst the earliest symptoms of the disease, preceding the motor manifestations by up to 9 years. Computed tomography (CT) images can be used to detect vocal fold motion changes related to these changes in voice. The anatomical position of the vocal folds is affected by the patient position, motion, phonation and breathing during scanning. Currently, radiologists manually adjust the vocal fold plane to analyse vocal cord changes manually. The large amount of modern CT imaging data capturing vocal fold plane makes it laborious and time-consuming. Localizing the vocal fold plane is a complex task and no known research automates the vocal fold plane extraction. Previously, we also worked to track tumour using PET images. |
|
Collaborators: Monash Medical Centre, Australia; Monash University, Australia; Peter MacCallum Cancer Institute, Australia; Royal Brisbane Hospital, Australia; Outcomes: Automated image segmentation and tracking algorithms from biomedical images. Publications:
|
Stroke Detection:Is one of leading cause of disability and mortality. An estimated 15 million people suffer from stroke incidence and around 6 million deaths every year. In 2015, an estimated 440,000 Australians and 1.6 million Indians suffered from stroke. Early detection of stroke is important to save people’s lives as late treatment will permanently damage brain causing death. Around 80% of cases are because of blood clots in brain. Administration of a thrombolytic medication to unblock the arteries after immediate onset (within 24 hours) of stroke is critical. However, administering thrombolytic agents require continuous monitoring. The first 24 hours after the first onset stroke has to be carefully monitored as stroke-like activities manifests and lack of continuous monitoring may lead to lost opportunities in timely treatments. In collaboration with Royal Melbourne Hospital, we have developed a proof-of-concept device based on accelerometer based on motor activity. |
|
Collaborators: Royal Melbourne Hospital, Australia Outcomes: An algorithm based on accelerometry signals to detect affected arm and provide NIHSS score. A Stroke Watch has been designed to estimate the NIHSS score. Publications:
|
Epileptic Seizure Detection:Epilepsy is one of the most common neurological disorders. It is affecting about 50 million people worldwide. Patients with epilepsy are known to suffer from recurrent unprovoked seizures. One of the challenges with epilepsy is the unpredictability of the seizures. Seizures may cause significant risk on prolonged events and require emergency medical attention. We have developed a system that uses wrist-worn accelerometer. We have developed an algorithm to detect non-epileptic seizures using accelerometer data. The system uses machine learning to classify the seizures. This was developed in collaboration with Royal Melbourne Hospital. |
|
Collaborators: Royal Melbourne Hospital, Australia Outcomes: Algorithms to detect convulsive and pseudo seizures based on unique accelerometry signatures. Publications:
|
Smartphone-based Low-cost Pulse Oximeter: Pulse oximetry is used in every developed-country clinic as a simple, painless way of measuring how much oxygen is in a patient's blood. Blood flowing into peripheral parts of the body (such as a fingertip or ear lobe) normally has almost 100% of the haemoglobin oxygenated, but in diseases such as pneumonia the "oxygen saturation" falls. An oximeter essentially consists of two LEDs, one emitting light in the red part of the spectrum and the other in the infrared, and a photo diode. However, despite their potential usefulness, oximeters are expensive (e.g. around $500) and rarely seen in developing countries. Cell phones are becoming increasingly common in developing countries such as Mozambique. The aim is to create a series of applications for cell phones as a platform for healthcare in remote and underserved communities in developing countries, using Mozambique as the site for development and testing. |
|
Collaborators and Partners: Microsoft; Nossal Institute for Global Health, Department of Information Systems, University of Melbourne Outcomes: Designed prototype of a low-cost oximeter sensor and it is currently under clinical validation trials at western hospital, Victoria. It can connect to mobile phones to provide information about measured sensor values. We are examining applications for Australian remote communities. |
Human Gait Pattern Analysis and Modelling:Understanding the underlying mechanisms and associated deficits in movement dynamics across the lifespan and the effects of pathological conditions, such as falls, will lead to many applications in the design and evaluation of diagnostic and assessment methods for human movement. Preliminary results obtained by our group have established the viability of the project for the "diagnosis" of gait instability and pathology. While in this application the experimental focus is both normal and ageing-influenced gaits, the outcomes are broad and generic having applications to other pathologies, and to rehabilitation, robotics and human performance. This work has been in collaboration with Victoria University Human Movement Lab. A few companies including SenSen Networks have expressed interest in such a technology apart from a few hospitals. |
|
Outcomes: Developed SVM-based machine learning models to detect and classify gait patterns. Publications:
|
Sleep Disordered Breathing:People with sleep disordered breathing (SDB) die prematurely with cardiovascular disease more often than someone without SDB. Early diagnosis and treatment of SDB could prevent adverse consequences. This investigation may provide essential information for introducing a novel screening device that can aid sleep specialist. This in turn may help prioritize patients, so that those in greatest need of treatment will undergo full PSG recordings in a timely manner, while those without apnoea will be able to avoid this tedious procedure. The work is supported by ARC Linkage grant scheme with Compumedics Ltd as the industry partner. |
|
Grants: ARC Linkage Projects (LP0454378) "New Techniques for Modelling, Diagnosis and Counter Measures for Cardiac Related Sleep Disordered Breathing" Collaborator: Compumedics Ltd Outcomes: We investigated into different machine learning techniques [support vector machines (SVM), Neural Network (NN), Quadratic discriminant (QD) model ] with an aim to find an appropriate model for the automatic detection of SAHS types from their respective overnight ECG recordings and estimation of relative degree of sleep disordered breathing. Publications:
|
Urban Noise Monitoring:Excessive noise could potentially be harmful. Some of the frequencies are known to cause increased disturbances among people with autism. We investigated the problem to provide a solution based on wireless networks. The solution offers city councils to understand dynamic noise pollution profile, assess its impact on health and wellbeing, and better plan for noise reduction and desirable urban sound-scape. This solution was developed in collaboration with the City of Melbourne. |
|
Contract: Noise Mapping: Designing an Urban Information Architecture to Record and Map Noise Pollution. Partner: City of Melbourne Outcomes: We proposed a frameowrk to monitor noise in urban environments. We also eveloped low-cost noise monitoring hardware considering energy-efficiency of noise monitoring sensor network, and development of protocols for efficient collection of high-sample rate noise data from sensors deployed over large geographical regions. This also included developing methodologies for analysis and visualisation of the noise pollution profile. Publication:
Media:
|
Air Quality Monitoring - Particulate Matter (PM):Increased level of pollution in the atmosphere contributes to respiratory health problems in the population. Particulate matter (PM), which is a fluid mixture of solids and liquids, is arguably one of the most dangerous of all pollutants to cause a population health hazard. Atmospheric pollutants, such as gases and particulate matters (PM) pose a threat to human health. In particular, there has been a strong focus on particulate matter as it is a common pollutant to cause population health hazards, especially respiratory illness. Monitoring of this pollutant is currently attained at low spatial resolutions due to the cost of accurate sensing devices. Even though these devices are highly accurate, given the distance they are placed apart from each other, the relevance of their measurements to an unmeasured spatial location in between sensors will be very low, which causes large estimation errors. |
|
Grants: ARC Linkage Projects (LP120100529); ARC LIEF (LE120100129); EU FP7 SocioTal and H2020 Organicity Outcomes: We developed wireless sensor nodes equipped with low-cost PM sensors to supplement the existing high-accuracy PM devices to improve the estimation at higher spatial and temporal resolutions. Furthermore, we designed spatiotemporal estimation algorithms and tools to efficiently estimate the data at unobserved locations using the combination of high-cost and low-cost sensors. Publications:
|
Environmental Monitoring:Great Barrier Reef Ocean Observing System (GBROOS): The Great Barrier Reef (GBR) consists of 3200 coral reefs extended over 280,000km2. Understanding the patterns of thermal stress and other environmental parameters is essential for monitoring the health of the coral reefs (eg., refer [10] for environmental change and impacts on GBR). The health of the coral reefs can be affected by cold water intrusions, hot water intrusions, coral calcification, ocean acidification, coral-algae phase shifts and the spread of coral diseases due to temperature increases. One of the GBROOS aim is to install sensor networks at seven reef sites along the GBR and monitor the reefs. Smart Environmental Monitoring and Analysis Technologies (SEMAT): The project aims to progress a working proof-of-concept next generation wireless marine sensor network to a commercial prototype to enable real time monitoring of the marine environment. The system will utilise an integrated, sensor network solution using underwater radio-frequency communication. Sensor Map: Recent developments in technology together with widely observed climate change phenomena have revealed coral reef ecosystems as critical areas greatly susceptible to impact of global climate variations as well as other man-made influences, but also as early indicators of such events. The need to understand and protect such delicate ecosystems has created an urgent demand for the sensor networks technologies to be deployed in order to perform essential environmental monitoring and information collection. The project has created a facility to publish the sensed data (visual, temperature, light, etc.) directly on the World Wide Web. The online data published periodically can be accessed on MS SensorMap webpage from anywhere. |
|
GBROOS - Partners and Collaborators: Australian Institute of Marine Science (AIMS); James Cook University; University of Queensland; SEMAT - Partners and Collaborators: University of Queensland, James Cook University, Queensland Cyber Infrastructure Foundation, Torino Foundation, Politechnic di Milan and the Danish Hydraulics Group Australia SensorMap -Partners and Collaborators: Microsoft funded project (SensorMap for Great Barrier Reef); Australian Institute of Marine Science (AIMS); James Cook University Outcomes: Developed sophisticated algorithms for modelling the effects of pollution in the Great Barrier Reef (GBR) from the deployed sensor networks. Our algorithms were able to detect the Cyclone Hamish (formed in the northern part of Australia and tracked through the main part of the reef) on the 9th of March 2009 using the sensor network stations at Heron and One Tree Islands. Publications:
Additional Links: |
BigNet Testbed:The development of a world class national research and industry capability in sensor network technology is of great national importance. Although Wireless Sensor Networks (WSNs) are being studied on a global scale, current research is still focusing on simulations and small-scale experiments. In particular for sensor networks, which have to deal with very stringent resource limitations and that are exposed to severe physical conditions, real experiments with real applications are essential. The effectiveness of simulation studies is severely limited in terms of the lack of scalability of network simulators for large networks, as well as the difficulty in modelling the complexities of the radio environment, power consumption on sensors, and the interactions between the physical, network and application layers. Therefore, we propose to face the real world, extending the simulations and laboratory experiments to large scale, real applications. Only by doing so will WSN research: (1) get an insight into the real needs of the users of sensor network; (2) face the world of errors, incomplete information, dynamics, etc. that is very hard to model; (3) perform large-scale experiments of several hundreds of sensor nodes. In the BigNet project, we intend to use this experimental facility as a testbed to demonstrate the algorithms developed on a large scale, facing real world problems, and developing solutions for a variety of applications |
|
Grant:ARC LIEF Grant (LE0883073) - "Bignet - A Distributed Wireless Sensor Network Testbed" Collaborators: The University of Melbourne; James Cook University; Deakin University Outcomes - deployment projects:
Additional Link: |