intelligent systems group (isg) - infotech oulu · 2009-05-18 · oulu institute, and the oulu...

INFOTECH OULU Annual Report 2008 29

INTELLIGENT SYSTEMS GROUP (ISG)

Professor Juha Röning, Professor Jukka Riekki and Professor Tapio Seppänen,Computer Engineering Laboratory, Department of Electrical and Information Engineering,

University of [email protected], [email protected], [email protected]

http://www.ee.oulu.fi/research/isg

Background and MissionThe Intelligent Systems Group’s (ISG) mission is to carryout long-term research on novel technologies and applica-tions of intelligent systems. The main objective is to de-velop enhanced adaptivity and context-awareness for smartenvironments. The research specifically focuses on the crea-tion of dynamic models that enable monitoring, diagnos-tics, prediction and control of target systems (living andartificial) or operating environments. It is our aim to makethe environment adapt to the users, instead of making theusers adapt to an inflexible environment. We believe thatby creating these novel components for smart environments,important enabling functionality will emerge that will mul-tiply the versatility and applicability of such living envi-ronments.

We see behaviour modelling as a major challenge in devel-oping truly intelligent and proactive environments. Humanusers of smart environments often behave in such a com-plex manner that it is hard to predefine and pre-program allof their behavioural patterns in the software. Models of userbehaviour are required that are able to grasp the user’s con-text at any moment, and to enable adaptation of the func-tionality of the intelligent environment to the situation athand. Further, it is essential to model the behaviour of thedevices controlled by the intelligent environment, as thisenables adaptation to environmental changes without re-programming. Systems should eventually learn and adaptautomatically through these models to perform their dutieseffectively.

Our research group combines a variety of key skills andtechnologies to aim at this goal. We have experience in thefollowing key technologies: system architectures and im-plementation of context-aware systems; modelling and rec-ognition of contexts from sensor signals; data miningalgorithms; learning nomadic robots; embedded systemstechnologies; software security; and smart environmentimplementations. The key application areas are: smart liv-ing environments in homes and institutes; industrial auto-mation; mobile robots; context-aware mobile devices; andwellness and medical applications. Each of these domainspossesses special characteristics, but, from the point of viewof developing algorithms for an intelligent system, they alsopossess remarkable similarities. They all produce a multi-tude of signals that represent the status of the system. Thetarget system behaviour should be modelled and recognizedfrom the signals. The application service should then actaccordingly. The availability of several application domains

yields many advantages: a solution to a special problem inone domain may offer added-value functionality in someother domain; our solutions are deployed by many of ourclient industries; solutions to a wide range of real-worldproblems define a credible and versatile tool-box that has amajor impact on our development-oriented sub-contractualprojects.

The group co-operates with many international and domes-tic partners. In applied research, the group is active in Eu-ropean projects. The group hosted three large Europeanproject meetings (ROBOSWARM, XPRESS, eConfidential)in 2008. In addition, several joint projects are funded bythe Finnish Funding Agency for Technology and Innova-tion (Tekes) and industry.

The group and its members are active in the scientific com-munity. For example, Prof. Juha Röning co-chaired numer-ous international workshops in the software security area:The 9th Winter School of the European Intensive Pro-gramme on Information Security Management and Tech-nology (IPICS) was arranged in Rovaniemi, and the 4thCrisis Management Workshop (CRM 2008) in Oulu, Fin-land. In information security, the group acted as a memberof the SAFECode International Board of Advisors, 2008.The group also coorganized a Vulnerability Prevention andSoftware Security seminar with a keynote speech by Mr.Howard Schmidt in April, 2008.

Prof. Tapio Seppänen is a leading figure in both the WellTechOulu Institute, and the Oulu School of Biomedical Engi-neering, an umbrella organization of wellness and medicaltechnology education at the University of Oulu and the OuluUniversity of Applied Sciences.

Several members of the group were also on the committeesof international conferences. The group’s expertise is rec-ognized, testified by the many invited talks and lecturesthat have been given. The Intelligent Systems Group hascommunicated its research to the public and its researchareas have attracted interest in the media.

Scientific ProgressIn 2008, the research at ISG concentrated on prototypingsmart environments, mobile and context-aware systems, datamining methods, signal analysis and secure programming.These were applied in context-aware mobile systems, in-telligent service robots, quality control of steel plant andspot welding processes, and analysis of biomedical ECGand EEG signals.

30 INFOTECH OULU Annual Report 2008

Two-dimensional magnetic map in a corridor of the Com-puter Engineering Laboratory, and the measurement sys-tem used in global self-localization experiments.

Research on Prototyping: from a Smart En-vironment towards Remote Distributed Intel-ligenceVerification of the developed methods and models in pro-totypes will be an important part of the research. To sup-port this activity, we will develop software and hardwarearchitectures for smart environments. In addition to verifi-cation, prototypes speed up the commercialization of theresearch results. In prototyping, we have set and tackledthe following objectives:

Developing hardware architecture for a smart environment.The aim is to develop a basis for devices operating in asmart environment. Modular technologies have been thesubject of our research for some years now. Modularity hasbeen achieved using a modular electronic concept, theEmbedded Object Concept (EOC) Atomi and a modularsoftware architecture - Property Service -, which providesan interface for these resources, and an opportunity to con-trol different robots and other devices of a smart environ-ment.

EOC is a concept that utilizes common object-orientedmethods used in software by applying them to combinedLego-like software-hardware entities. These modular enti-ties represent objects in object-oriented design methods,and they function as the building blocks of embedded sys-tems.

This concept enables one to build new embedded systemsfrom electronic Lego-like building blocks. The goal of theEOC is to make designing of embedded systems faster andeasier while preserving the commercial applicability of theresulting device. The EOC enables people without com-prehensive knowledge in electronics design to create newembedded systems. For experts, it shortens the design timeof new embedded systems. Implementing the conceptualidea of embedded objects has been successfully imple-mented with the Atomi II framework.

The EOC research has proceeded by developing new ob-jects. The focus of the research has been on improving theirrobustness, and testing the concept in practice by support-ing other research projects. The research will continue to-wards user friendly development tools in order to furtherimprove the concept.

Research on Mobile RoboticsLocalization is one of the fundamental problems in mobilerobotics, as in many applications a robot needs to know itslocation in order to perform its tasks. A global self-locali-zation technique was proposed that utilizes observations ofthe ambient magnetic field. This study was inspired by evi-dence that animals use the magnetic field of the Earth fortrue navigation. The experiments reported in this articlesuggest that (especially) modern buildings with reinforcedconcrete and steel structures have unique spatially varyingambient magnetic fields that can be used for navigation, invery much the same way as the Earth’s magnetic field, buton a smaller scale. In principle, a non-uniform ambientmagnetic field produces different observations, dependingon the path taken through it. The approach provides a prom-

ising and simple technique for solving the global indoorself-localization problem.

The current technique is applicable for a one-dimensionallocalization problem, i.e., for localizing a robot or a personwithin corridors. However, the proposed approach couldbe extended to two- or three-dimensional localization prob-lems, assuming that maps can be provided. In some appli-cations, the proposed approach may provide an alternativeto machine vision based approaches, especially when onlyone-dimensional localization is needed, or when the illu-mination of the environment changes. On the other hand,the proposed technique may also be used in parallel withmachine vision and range finder based approaches in orderto overcome possible sensor aliasing problems.

The experiments suggest that the ambient magnetic fieldmay remain sufficiently stable for longer periods of time.In the conducted experiments, the magnetic field remainednearly unchanged, although some variations were observed.These variations did not, however, have a great impact onthe localization performance. On the other hand, the mag-netic field is not sensitive to many environmental changesthat may affect other localization techniques, such as vi-sion or range finder based techniques which rely on visualor geometrical features. For example, the magnetic field isnot sensitive to non-magnetic dynamic or static objects, norto changes in illumination.

Multi-robot exploration methods can be utilized to exploreunknown environments using a team of robots. In ISG, anovel multi-robot exploration algorithm has been studied.The exploration algorithm chooses frontier cells for each


From left to right: paths of the robots in a coordinated case; the occupancy map produced by a team of coordinatedrobots; paths of the robots in an uncoordinated case; and the corresponding occupancy map.

TACO generatedpaths for two, three,and four robots in ahospital environ-ment (left), and forthree robots in theComputer Engi-neering Laboratory(right).

On the left: an experimental en-vironment including 1) a ma-chine vision localization system,2) measurement area, 3) batterycharger and 4) central comput-ing unit. At the top middle andright are integrated meansquared errors for stationaryand non-stationary spatial mod-els as a function of samples forambient light and magnetic fluxdensity, respectively. At the bot-tom middle and right are regres-sion results after 50 samples forthese quantities.

individual robot so that the overall exploration time is mini-mized. The frontier cell is a cell between a known and anunknown area. Whenever a robot arrives at the frontier cell,new information is received around the cell. Each cell ofthe occupancy grid map contains a probability that the cellis occupied. The cell is occupied if the corresponding areain the environment is covered by an obstacle. The only re-quirement is that the map must allow the distinction be-tween known and unknown areas and it must compute travelcosts for the individual robots. The algorithm simultane-ously takes into account the cost of reaching a frontier celland its utility. The cost of the cell depends on the occu-pancy probabilities of the cells along the path of the robot,and the distance between the robot and the cell. Whenevera frontier cell is assigned to a robot, the utilities of the nearbycells are reduced. The utility of the frontier cell also de-pends on the number of robots moving near to the cell. Ourresults suggest that this coordinated multi-robot explora-tion technique is more effective than uncoordinated tech-niques, as is shown below.

Path optimization is a very important aspect of multi-robotsystems in which the available resources, such as batterylife, can be limited. Many real-world multi-robot path opti-mization problems can be restated as an instance of theMultiple Travelling Salesman Problem (MTSP). A novelant colony based algorithm called TACO was proposed forsolving the MTSP with a min-max objective. Its competi-tiveness was shown by comparison to neural network basedapproaches, and its feasibility in multi-robot systems wasdemonstrated in a simulation environment.

In autonomous environment modelling research we sensethe environment using mobile robots. This enables selec-tion of optimal sampling locations in order to provide amodel with maximum accuracy. In geostatistics studies,optimal spatial sampling has traditionally focused on theselection of sampling locations in advance. However, withmobile sensors, we are able to select the locations based onthe current model which increases the accuracy of the modeland decreases the measuring time.


A prototype of a robot boat for sensing aquatic environ-ments: 1) an embedded computer with a WLAN communi-cation interface, 2) sensors, and 3) GPS positioning sys-tem.

The gesture controller.

A user controlling an application on a wall display with ges-tures.

We have developed mobile sensing systems for variousenvironments, including indoor, outdoor, and aquatic envi-ronments. In indoor environments, the most important meas-ured quantities are temperature, humidity, and gasconcentrations. In aquatic environments, measurements ondissolved oxygen, pH levels, and temperature provide im-portant information for modelling these environments. Ourresearch integrates the latest results from geostatistics, andmulti-robot systems to enable implementation of fully au-tonomous, and adaptive sensing systems.

Research on Context-Aware ServicesContext-aware services adapt to the user’s situation. Wehave studied context recognition in several different appli-cation areas, including human physiology, wellness, urbancomputing, smart office spaces and body sensor networks.The aim is to find and develop methods for recognizing theuser’s context from sensor data using signal processing,pattern recognition and machine learning methods. Con-text can be used to identify the services that are relevant inthe situation at hand - and to adapt these services.

During the year 2008, the research on context recognitionhas continued with gesture recognition. We have developednovel solutions and applied Hidden Markov models to en-able gestural controlling with an ordinary mobile phone.The gestures are recognized from 3D acceleration meas-urements that the N95 or Nokia Sport 5500 can produce,but also other 3D acceleration sensor devices are applica-ble. With the gestures, different applications can be con-trolled, including the inner functionality of the mobile phone.The gesture controller is configured with (gesture, com-mand, application) triplets. When the controller recognizesa gesture, it sends the corresponding command to the speci-fied application. The system architecture is presented in thefollowing figure. The user can teach the gestures and asso-ciate each gesture to a controlling command. For example,a user can select a clockwise circle drawn in the air to cor-respond to the command for opening the calendar applica-tion on an N95. The second figure shows a user controllingwith gestures an application on a wall display. The soft-ware was published as open source on November 2008 andit can be downloaded from the group’s web site.

During the last year the group also studied recognition ofcontext from ambient audio. Here, audio from users’ eve-ryday environments was recorded using a mobile device,and through signal processing and classification, it was usedas a cue of the users’ current context. Part of the work wasdone in collaboration with Nokia Research Center and MIT,and funded by the Fulbright organization.

The group is strongly moving on to studying signal process-ing in networked sensing systems. The group members areparticipating the UbiCity project, where a tool on gather-ing, storing and processing sensor measurement from a het-erogeneous sensor network is being built. The group hasparticipated in organizing a data mining workshop onsensorwebs, databases and mining in networked sensingsystems (2nd SWDMNSS 2008) together with Japanese andGerman researchers. This workshop and several two-wayresearch visits with Tokyo Denki University’s UbiquitousNetworking Laboratory have created international collabo-ration.

In addition to recognizing a user’s context, sensor data canbe used to build a physical user interface in which a useruses a mobile device as a physical object rather as a tradi-


A physical user interface for controlling videos on a walldisplay.

tional I/O device. The device is equipped with sensors andthe actions of the user are recognized from the sensor dataand interpreted as commands. Gesture controlling is anexample of a physical user interface; the user commandsthe system by waving the terminal on the air. In addition togestures, the group is actively developing touch-based physi-cal user interfaces. These user interfaces are based on RFIDtechnology. An RFID tag storing service parameters isplaced behind an icon advertising a service. When a usertouches the icon with a mobile device that is equipped withan RFID reader, the service parameters are read from thetag and delivered to the system. Touch-based user inter-faces are an effective way of controlling a system, as theyproduce rich contextual information: who requests a serv-ice, which service, when, and where. With such interfaces,the user remains in control.

During 2008, the group developed several innovative touch-based user interfaces in co-operation with the University ofLapland. The group also developed further the REACHeSplatform that can be used to link physical user interfaces toInternet services. The figure below presents a touch-baseduser interface for controlling videos on a wall display. Atthe bottom left we see an icon that advertises a service inthe environment, next to a wall display. When a user touchesthis icon with a mobile device, a video is shown on the walldisplay (top of the figure) and a remote control UI is cre-ated on the mobile device’s display (bottom right). The us-ers can now control the video on the wall display usingtheir mobile device Two alternative physical user interfacesfor the same service have been developed: Control Paneland Control Cube.

In addition, we created an application for sharing informa-tion at our university’s Zoological Museum. In this appli-cation, icons are placed next to exhibition items, the stuffedanimals. Touching an icon with a mobile device brings a

list of item-related documents to the device’s display. Theuser can download the selected files to a device and forexample, listen the sound of an animal, or study a photo-graph. Over 300 students from local primary and second-ary schools tested the application in spring 2008.

Research on Data MiningBiosignal ProcessingCardiovascular signal processing. Sudden cardiac arrest isthe most common cause of death in western countries. Itaccounts for approximately 50% of cardiovascular deaths,and apparently has a highly variable pathophysiologicaletiology. The risk of sudden cardiac arrest is high in certainsubgroups of patients with a history of myocardial infarc-tion and depressed left ventricular function. The key ques-tion in research is why do some subjects develop ventricularfibrillation during acute coronary occlusion, while otherssurvive this episode without fatal arrhythmia. The challengefor research is to develop approaches or techniques thatwill allow the screening of the specific risk for fatal ven-tricular arrhythmias as a predictor of the first event in pa-tient populations that have a low cumulative risk, butgenerate a large number of victims. In addition, the predic-tive value of many known risk factors of sudden cardiacarrest among patients with known heart disease has not beendefinitively established.

Invariant trajectory classification of dynamical systems witha case study on ECG. An invariant pattern recognition frame-work for classification of phase space trajectories of non-linear dynamical systems was developed. Using statisticalshape theory, known external influences can be discrimi-nated from true changes of the system. The external effectsare modelled as a transformation group acting on the phasespace, and variation of the trajectories not explained by thetransformations is accounted for using principal componentanalysis. The approach suggested is highly adaptable to awide range of situations and individual differences. Themethodology presented is applied to detect abnormalitiesin electrocardiograms. Results based on measured data in-dicate that the model developed is resistant to the effects ofrespiration and body position changes, which are abundantin ambulatory conditions and cause significant morphologi-cal artifacts in the signal. The results also show that thedetection of an artificially induced acute myocardial inf-arction is achieved with high performance. Due to its lowcomputational complexity, the method developed can beimplemented in real-time. This method also adapts to mor-phological changes caused by various heart conditions.

A method for estimating the severity of myocardial infarc-tion from multi-channel ECG. The measurement of infarctseverity is an important element in the overall care of pa-tients with STEMI (myocardial infarction with ST segmentelevation) patients. The severity of the MI (myocardial in-farction) is commonly estimated by cardiac biomarker meth-ods and echocardiography. We developed a method forestimating the severity of the MI by combining an actionpotential based computer model and 12 lead ECG patientdata. The estimators of the severity of the anterior and infe-rior MI were developed using an EDL (equivalent double


layer) model. The best combinations of the single param-eter based estimators were found by using multiple linearregression analysis. The correlations between the final esti-mators and two clinical estimators of the severity of the MIwere calculated. The severity estimators correlated to themaximum troponin value with r value 0.615 and to the ejec-tion fraction with r value 0.428. On the grounds of the re-sults, it is possible to calculate a coarse estimate for themaximum troponin value, and therefore the severity of theMI, from the standard 12 lead ECG by using the simulationmodel based method. The estimated troponin value can beused to yield a fast assessment of tissue damage of an am-bulatory cardiac patient that is suspected to have an infarc-tion.

Improving Reliability of ‘Total-Cosine-R-to T’ (TCRT) inPatients with Acute Myocardial Infarction. The parameterTCRT (Total Cosine R-to-T) calculated from ECG record-ing has been shown to have a remarkable prognostic valueas a predictor of the outcomes of coronary artery diseaseand acute myocardial infarction (AMI) patients. The TCRTis conventionally calculated using an algorithm producedby Acar et al. (1999). In this study, the reliability of theTCRT algorithm was tested with the ECG data of a healthygroup (n = 25) and the AMI group (n = 45). Typical prob-lems occurred in the detection of the maximum of the Tvector (9% of patients), the bounding of the R wave (18%),a comprehensive segmentation (11%), and a decreased con-gruence between TCRT and the spatial QRS-T angle (33%).The results show that small improvements to the basic al-gorithm can decrease the number of failures by up to 82%in AMI data. It is concluded that segmentation propertiesshould be improved in the basic TCRT algorithm in orderto maintain the diagnostic value of TCRT in different pa-tient data.

EEG signal processing. The aim of this task is to developnew methods and algorithms for estimating the depth ofanaesthesia from a multi-channel EEG. A new concept ofrelative induction time was recently developed by our group,which enables a significant improvement in the depth esti-mation from EEG. The new method produces a time-con-tinuous value for the depth estimate and accurately predictsthe instant of loss of consciousness. The method was ex-tended so that it can model the depth of anaesthesia to sucha depth that burst-suppression of EEG starts to evolve. Theteam has so far focused on describing the spectral charac-teristics of EEG related to the induction of anaesthesia withpropofol. The time-frequency characteristics of EEG thatare consistent between individuals during induction of an-aesthesia were determined. The relationship between theEEG frequency progression pattern and the clinical end-points, such as loss of obeying verbal command, was shown.A mathematical method was developed for describing theconsistent EEG frequency progression pattern during an-aesthesia. The method has provided us the possibility tostudy in detail the effects of analgesic drug (remifentanil)on the EEG-based depth of anaesthesia estimation. The re-sults indicate that remifentanil modifies the relationship ofEEG spectral changes and clinical endpoints in propofolanaesthesia. The effects of remifentanil on the onset of burstsuppression pattern were also presented.

Text production speedup technologies. The feasibility ofword prediction was studied in a highly inflected language,Finnish, where words are used in many case forms, a topicseldom addressed in the context of word prediction. Sinceabout one third of words will appear in uninflected formsin Finnish, simple prediction methods, e.g. word comple-tion, typically employed in uninflected languages such asEnglish can be used as such in the prediction of uninflectedwords. The preliminary results obtained show that about45% of characters can roughly be saved in Finnish wordprediction in general for uninflected words. Secondly, theutility of predicting entire phrases instead of single words,(as usually done in word prediction), consisting of two ormore words was investigated in English word prediction.The results obtained show that about 70% of characters ofthe phrases included in a lexicon of some 7,000 phrases oflegal English could be saved in theory when using the bestsearch key for their prediction.

Data Mining SystemsThe focus areas of data mining research were re-organisedduring the year 2008, and the research is now carried outwith a quite unique approach. The research challenges aredivided into three mutually supportive categories; the re-search of algorithms producing knowledge, software run-ning the algorithms and knowledge bases storing theacquired knowledge. When put together, these three cat-egories form a strong combination which can be applied tovirtually any phenomena where data can be processed intoknowledge.

The research on algorithms was focused on advancing meth-ods for time series analysis, variance modelling and nov-elty detection. Special interest was directed to establishingdata driven methods for these areas, where the exact shapeand nature of the observed data can be used to characterizethe phenomena under study. In software research, the im-plementation of the first version of new software architec-ture for running the algorithms came to successfulconclusion. The architecture presents the algorithms as in-formation generating logical devices to which the informa-tion measuring physical devices connect. A suitablecombination of logical and physical devices can thereafterbe used to form a data mining software application. In

EEG activity in different frequencies during induction ofpropofol anaesthesia.


Three focus areas of the data mining research; algorithmsgenerating knowledge, software running the algorithms andknowledge bases storing the acquired knowledge. In 2008, data mining methods were applied for example to

monitoring tool usage of associates in production lines. Themethod could be used to improve worker ergonomics.

knowledge base research, the general ideas behind the roleof knowledge bases were established, and the implementa-tion of the first version of a concrete knowledge base wascarried to a successful end.

The research has focused on three projects, XPRESS (2007–2010), SAMURAI (2006–2008) and MIDAS (2008–2010).XPRESS is an integrated (IP) EU-project, where new ap-proaches for managing and optimizing the operation of en-tire production facilities are being developed. In 2008, datamining algorithms were created, for example, for recogniz-ing the tools the associates in the production lines are usingat a particular time. The analysis is based on accelerometerdata, and it can help, for example, in finding more ergo-nomic ways of working. The software architecture andknowledge bases were applied to storing the generatedknowledge in this application and also a wider range ofapplications in factories.

The SAMURAI-project was completed during 2008, andthe results were evaluated to be so good that they gave riseto the starting and granting of new funding for a project,MIDAS, which is based on the results. In both projects,methods for utilizing data originating from a steel factoryand human being were developed. The algorithms appliedto steel production helped in reducing the variability of thequality of steel plates. Also, methods for pointing out theright time to perform a production model parameter updatewere reported. In the analysis of the human data, methodsfor estimating the energy expenditure of various sports ex-ercises based on accelerometer data were reported for thefirst time. Also, research for recognizing sport activities fromaccelerometer time series was continued successfully.

The group had collaboration in the focus area of data min-ing both internationally and nationally. The researchers gavepresentations of their results in 10 conference presentationsaround Europe, the USA and Asia. Co-operation with Eu-ropean researchers and industry was intensive also thanksto the XPRESS project in which 17 different European or-ganizations are participating. The data mining group also

belongs to the Centre of Advanced Steel Research, whichis founded to gather together from both at home and abroadsteel research expertise at the University of Oulu.

At the end 2008, the future for data mining research seemsbright, and major advancements in all of the three focusareas (algorithms, software and knowledge bases) can beexpected in 2009. International collaboration can be ex-pected to deepen even more during the next year thanks toforthcoming long term research visits established in 2008.

Research on Software SecurityWithin the Intelligent Systems Group, the Oulu UniversitySecure Programming Group (OUSPG) has continued re-search in the field of implementation level security issuesand software security testing. Software implementation mayintroduce potential for unanticipated and undesired programbehaviour, e.g. an intruder can exploit the vulnerability tocompromise the computer system.

In 2008, the research focus at OUSPG remained on black-box methodology for improving software security. OUSPGapproaches the problem from three different directions,namely, network traffic data-mining and visualization, pro-tocol genes and protocol dependence.

Causal relationships. The aim is to develop methods forinferring causal relationships in complex systems. The re-search applied data mining to network traffic to find datarelevant to the system being analysed. Applications of themethod include getting an overall view of the communica-tion patterns of complex systems, diagnostics and securityrisk assessment.

Information network environments are more complex thanever before, and the complexity will increase in the future.One important factor affecting the increase in the complex-ity is the convergence of information networks. One seem-ingly simple event may generate a number of small actionsin different parts of the network. Additionally, differentcomponents may use varying protocols. Thus the securedevelopment, deployment and management of complexnetworks is laborious and requires in-depth understandingof different fields.


Practical RFID security research in progress.

Structural inference methods used in plagiarism detection.

Network traffic data mining research applies black-boxmethodology in understanding the behaviour of differentcomponents in computer networks. Examples of systemswhere this work has been applied are operator WLANs,firewalls and malware. Operator WLANs are a good exam-ple of complexity due to the large amount of infrastructureinvolved (e.g. integration with traditional phone networkauthentication infrastructure). Firewalls may have uninten-tional information leaks due to misconfiguration or imple-mentation problems. Malware is software that is purposelyobfuscated to make reverse- engineering and removal diffi-cult. By monitoring the external behaviour of malware, it ispossible to understand how it functions.

Identification of protocol genes. This research, PROTOS-GENOME, approaches the problem of complexity from theother direction, by developing tools and techniques for re-verse-engineering and identification of protocols based onusing protocol genes - the basic building blocks of protocols.The approach is to use techniques developed forbioinformatics and artificial intelligence. Samples ofprotocols and file formats are used to infer structure fromthe data. This structural information can then be used toeffectively create large numbers of test cases for this proto-col.

The PROTOS-GENOME project produced test suites forseveral archive file formats that are typically implemented,for example, in anti-virus products. The test suites havebeen distributed to vendors in cooperation with FICORAand NISCC. A number of vulnerabilities was found in widelyused antivirus software products. The method has been alsoapplied to find plagiarism in text.

As a continuation of OUSPG’s practical security research,work was conducted on the security of RFID systems. Thiswork is on-going in 2009.

Vulnerability management of the information infrastruc-ture contributes to protocol dependence. Another angle onbattling complexity is through technology dependencies.This activity studies the impact factor of different technolo-gies on CNI, and develops a visual model for understand-ing dependencies related to protocols. This wasaccomplished by extending traditional Wikis, which are

effective mass collaborative authoring services withgraphing extensions. Graphingwiki enables deepened analy-sis of the Wiki data by augmenting it with semantic data ina simple, practical and easy-to-use manner. Visualisationtools are used to clarify the resulting body of knowledge sothat only the data essential for a usage scenario is displayed.Logic inference rules can be applied to the data to performautomated reasoning based on the data. Perceiving depend-encies among network protocols presents an example usecase of the framework.

A newly established area of research in OUSPG is compu-ter-assisted teaching. This work originated in 2005 withRAIPPA, software for automated software for programminglaboratory pre-exercises. Work on RAIPPA continued in2008 with funding from Campus Futurus, an organizationwithin the university that promotes the use of ICT in teach-ing. The system has improved learning results significantly,and it is currently being piloted with courses from through-out the university.

Exploitation of ResultsThe results of our research were applied to real-world prob-lems in many projects, often in collaboration with indus-trial and other partners. Some examples of exploitation aredescribed below. Especially during the reported year, out-door robotics was a new area for exploitation of our re-search results.

The Intelligent Systems Group utilizes a robotics labora-tory and pressure-sensitive floor (EMFi material) installedin our laboratory as part of a smart living room. Other equip-ment includes a home theatre, two degree-of-freedom ac-tive cameras, four mobile robots and one manipulator, aWLAN network, and various mobile devices (PDAs, a tab-let PC, Symbian mobile phones). WLAN positioning cov-ers a large part of the campus (including the laboratory),and a home automation network is being installed. Our aimis to gradually build a versatile infrastructure that offersvarious generic services for pervasive applications. Natu-rally, this kind of environment enables realistic experimentsthat lead to a better understanding of such applications.


Micromanipulation platform.

The embedded objects implemented according to the AtomiII Framework specification are called Atomi objects. Sev-eral different Atomi objects have been created for real lifetests. The Atomi objects have been used in several projects,and they have proved to be very usable.

The Atomi objects are being used both in pure researchprojects and in projects that aim ultimately at commercialproducts. The Atomi objects have been applied totelepresence robots, nanoscale manipulation and measure-ment technology, a hand held medical device, and severalrobot applications. The most recent applications include,for example, a micromanipulation platform and a bio filmmeasurement device.

The micromanipulation platform (shown below) is a de-vice platform that can be used for different applicationsthat require actuation and sensing in nanometer resolution.Presently, nanoactuation devices on the market are very ex-pensive, and often limited in applications. This platform ismostly built with off-the-shelf components and Atomi ob-jects, and thus brings about a reasonable cost for the instru-ment.

The device is based on a generalized modular architecturewhich covers both device hardware and the control soft-ware in a PC. The modular architecture enables a swiftchanging of the actuators, sensors and tools with minimaleffort and reusable source code, thus being an ideal framefor various applications.

The platform consists of a haptic 3D controller, a piezo-electric actuator device, a probe and a PC. The haptic 3Dcontroller enables manual control over the robot for theuser. The controller has a pen-like handle which can bemoved with six dimensions of freedom. The pen has twobuttons which can be assigned to different functions. Thecontroller also provides a haptic feedback option that canbe used to let the user feel what the robot feels.

The piezoelectric actuator device consists of three linearpiezo stages with position encoders. This actuator movesthe measurement. The device can be easily changed. Wehave used a SmarAct piezo actuator, which can move in aresolution of 5 nm in 3 cm range.

The probe (measurement head) is connected to the actua-

tor. It can be customized to every device separately. In ourtest case, the probe is a silicon strain-gauge force sensorAE-801, which is attached to mechanically custom designedarm. The probe is connected to a voltage amplifier, andfurther to AD-conversion on 24-bit precision AD converterAtomi with USB and Power Atomi objects. A specially de-signed needle is attached to the force sensor, to feel thesurroundings. In some applications the probe can includean additional actuator. This actuator can be, for example, amicro gripper, which also can be controlled via Atomi ob-jects.

A bio film measurement device measures the resistance of athin soft coating i.e. the bio film. This device is used inmeasuring the properties of different type of coatings forcertain purposes. The thickness of the coating varies be-tween 1 µm and 2 mm. During the measurement, the coat-ing is on a level surface in a bowl filled with water. Theresistance is measured between a probe needle and the sur-face that the coating is located on. The surface is a planemade of material that conducts electricity. The needle is10 µm in diameter.

The measurement is not simple. In order not to damage thebio film, the measurement current must be very low and itspolarity must constantly change. Otherwise the sensitivesurface may burn or otherwise get damaged. Furthermore,the measurement must be made with several points on thebio film in a reasonable amount of time. These require-ments demand an intelligent measuring method and thusthe programmable Atomi objects are very well suited forthe job.

The next figure shows the setup of the device. It consists ofseveral Atomi objects, the piezo actuators, the measurementprobe and a PC for the high level control and a UI for thesystem. The Atomi objects that are in this device includetwo PiezoLegs Atomi objects, a USB Atomi, a power Atomiand an AD-DA converter measurement Atomi.

The results of the multi-robot and distributed sensing re-search will be utilized in a real world application scenarioas a part of the ROBOSWARM project, which is an EUproject that was launched in November 2006. Together witheight other participants, the University of Oulu has an im-portant role in this project, which aims to develop an openknowledge environment for self-configurable, low-cost androbust robot swarms usable in everyday applications. Ad-vances in the state-of-the art of networked robotics are pro-posed through the introduction of a local and globalknowledge base for ad hoc communication within a low-cost swarm of autonomous robots operating in the surround-ing smart IT infrastructure. For the ROBOSWARM projectISG has developed a custom embedded control system forthe swarm of robots. The Sensor and Connectivity Board(SCB) was designed with the aim of provide a modular ro-bot platform for swarm robotics research and applicationdevelopment, where simplicity and modularity are key-fac-tors. The SCB is used to integrate all sensor modalities andthe motor control into one logical device, which can beaccessed through a USB port from any computer havingUSB host capability. This solution providing easy access toall sensors, and flexibility for selecting an embedded host


An instrumented Robotic Team Member (RTM) built for theROBOSWARM project.

Mörri platform performing in European Space Agency'sLunar Robot Challenge in Teide, Tenerife.

computer to perform the high-level data processing, andmotion control computing. The sensor set currently sup-ported by the SCB includes an RFID reader, sonars, a laserrange finder, 3-axis accelerometer, gyroscope, photo diode,thermometer, humidity and atmospheric pressure sensors,3-axis magnetometer, and a real-time clock. SCB also fea-tures a built-in interface to iRobot’s Create mobile robot,but it can easily be connected to any robot platform withthe UART connection. In addition, SCB provides I2C,RS232, and SPI interfaces for adding new external devicesto expand the functionalities of the SCB even further.

During the reporting year, the group continued utilizingoutdoor robotic systems. Development and utilization ofMörri, a multipurpose, high performance robot platformcontinued, and major components for software architecturewere implemented. The software architecture further de-veloped the earlier work of Property Service Architecture.The main focus of development has been on multi-purposecontrol architecture, that can be used to integrate variousalgorithms, methods and sensors to one real-time system.Development of the robot included platform mechanic de-sign, high power brushless motor control electronics andsoftware, sensor integration, robot world modelling androute reasoning user interfaces. The user interface is de-signed for field use, including wearable control devices,and use of Google Earth software for visualizing and con-trolling the robot’s global path.

The Group participated in the Military European Land Ro-bot Trial 2008, held in Hammelburg Germany in July. M-Elrob is the biggest outdoor robot event in Europe, andparticipants are research facilities and companies that rep-resent the state-of-the-art in Europe in this research area.From four scenarios of competition, Mörri succeeded intwo, by winning the Camp Security scenario, and gettingfourth place in the Mule transportation scenario. In Octo-ber, the same Mörri platform was used in the EuropeanSpace Agency’s Lunar Robot Challenge (Esa-LRC) held inTeide, Tenerife. A new application module was imple-mented, including a sampling device for soil samples. Thetask in competition was to get samples from a 15 m deepcrater with up to a 40 degrees slope. Conditions in the chal-

lenge were temperature -2 degrees Celsius, hard wind andtotal darkness (without ambient light). Our robot was thirdin the competition and won world-wide publicity in news-papers, and on radio and TV.

Future GoalsWe will continue to strengthen our long term research andresearcher training. We will also continuously seek oppor-tunities for the exploitation of our research results by col-laborating with partners from industry and other researchinstitutions in national and international research programsand projects. The group is a founding member of the Euro-pean Robotic Network of Excellence (EURON). The groupis a contract member of EURON II which was approvedfor the EU’s FP6 as a Network of Excellence.

We will strengthen our international research co-operation.Within the last years, the group has created collaborationprojects with Japanese researchers. Several researchers havevisited Waseda University’s Distributed Computing Labo-ratory (DCL) to study ubiquitous computing paradigms. Thecollaboration has given rise to results on sentient (recog-nizing the user of an artefact), as well as activity recogni-tion and the corresponding architectural and data collectionissues. With the University of Tianjin, China we have a jointproject in which methods and a system will be developedfor vision-based navigation of Autonomous Ground Vehi-cles, which utilize an omni-directional camera system asthe vision sensor. The aim is to provide a robust platformthat can be utilized in both indoor and outdoor AGV (Au-tonomous Ground Vehicles) applications. This co-opera-tion will continue.

In the USA, we will co-operate with the Human-ComputerInteraction Institute in Carnegie Mellon University. A doc-toral student from ISG will make a one year research visitto the institute and co-operate with assistant professor AnindD. Key. The research will be on human modelling in thearea of human machine interaction.

A new co-operation agreement has been sealed during year2008 with IDIAP research in Switzerland. A post doctoralresearcher from ISG will make a one year research visitstarting March 2009. Shorter research visits to Europeanpartners of EC funded projects are also planned.


Personnel

srotcod&srosseforp 8

stnedutsetaudarg 72

srehto 24

latot 77

sraeynosrep 55

External Funding

ecruoS RUE

noitacudEfoyrtsiniM 000282

sekeT 000522

cilbupcitsemodrehto 00024

etavirpcitsemod 000451

lanoitanretni 000235

latot 0005321

Selected PublicationsAskola K, Puuperä R, Pietikäinen P, Eronen J, Laakso M, HalunenK & Röning J (2008) Vulnerability dependencies in antivirus soft-ware. Second International Conference on Emerging Security In-formation, Systems and Technologies (SECURWARE '08), CapEsterel, France, 273-278.Fujinami K & Riekki J (2008) A case study on an ambient displayas a persuasive medium for exercise awareness. Lecture Notes inComputer Science: Persuasive Technology, Springer, 5033: 266-269.Halunen K, Rikula P & Röning J (2008) On the security of VSHin password schemes. Third International Conference on Avail-ability, Reliability and Security (ARES 2008), Barcelona, Spain,828-833.Juutilainen I & Röning J (2008) Modelling conditional variancefunction in industrial data: A case study. Statistical Methodology,5(6): 564-575.Kemppainen A, Mäkelä T, Haverinen J & Röning J (2008) Anexperimental environment for optimal Spatial Sampling in a Multi-Robot System. Intelligent Autonomous Systems 10 (IAS-10),Baden Baden, Germany, 54-63.Koivikko M, Perkiömäki J, Karsikas M, Salmela P, Tapanainen J,Ruokonen A, Seppänen T & Huikuri H (2008) Effects of con-trolled hypoglycemia on cardiac repolarization in type I diabetes.Diabetologia, 51: 426-35.Kortelainen J, Koskinen M, Mustola S & Seppänen T (2008)Remifentanil modifies the relation of electroencephalographicspectral changes and clinical end points in propofol anesthesia.Anesthesiology, 109: 198-205.

Kortelainen J, Koskinen M, Mustola S & Seppänen T (2008) Time-frequency properties of electroencephalogram during inductionof anesthesia. Neuroscience Letters, 446: 70-4.Koskimäki H, Juutilainen I, Laurinen P & Röning J (2008) De-tection of Correct Moment to Model Update. Lecture Notes inElectrical Engineering: Informatics in Control, Automation andRobotics, Springer Berlin Heidelberg, 24: 87-94.Peltola M, Tulppo M, Kiviniemi A, Hautala A, Seppänen T, BarthelP, Bauer A, Schmidt G, Huikuri H & Mäkikallio T (2008) Respi-ratory sinus arrhythmia as a predictor of sudden cardiac deathafter myocardial infarction. Annals of Medicine, 40: 376-382.Pietikäinen P, Viide J & Röning J (2008) Exploiting Causalityand Communication Patterns in Network Data Analysis. 16th IEEEWorkshop on Local and Metropolitan Area Networks (LANMAN2008), Cluj-Napoca, Transylvania, Romania, 114-119.Riekki J, Sánchez I & Pyykkönen M (2008) Universal RemoteControl for the Smart World. Lecture Notes in Computer Sci-ence: Ubiquitous Intelligence and Computing, Springer, 5061:563-577.Röning J, Haverinen J, Kemppainen A, Mörsäri H & Vallivaara I(2008) Smart Systems for Distributed Sensing. Proceedings of11th Biennial Baltic Electronics Conference (BEC2008), Tallinn,Estonia, 21-30.Schaberreiter T, Wieser C, Sánchez I, Riekki J & Röning J (2008)An Enumeration of Rfid Related Threats. Proc. The Second Inter-national Conference on Mobile Ubiquitous Computing, Systems,Services and Technologies (UBICOMM’08), Valencia, Spain, 381-389.Seydou F, Duraiswami R & Seppänen T (2008) Numerical solu-tion of electromagnetic scattering by multiple cylinders. ACESJournal, 23.Siirtola P, Laurinen P & Röning J (2008) A Weighted DistanceMeasure for Calculating the Similarity of Sparsely DistributedTrajectories. Seventh International Conference on Machine Learn-ing and Applications, San Diego, USA, 802-807.Suutala J & Röning J (2008) Methods for Person Identificationon a Pressure-sensitive Floor: Experiments with Multiple Classi-fiers and Reject Option. Information Fusion Journal, Special Is-sue on Applications of Ensemble Methods 9(1): 21-40.Tamminen S, Juutilainen I & Röning J (2008) Product DesignModel for Impact Toughness Estimation in Steel Plate Manufac-turing. The International Joint Conference on Neural Networks(IJCNN 2008), Hong Kong, 990-993.Tiinanen S, Tulppo M & Seppänen T (2008) Reducing the effectof respiration in baroreflex sensitivity estimation with adaptivefiltering. IEEE Transactions on Biomedical Engineering 55: 51-9.Vallius T & Röning J (2008) Low Cost Arbitration Method forArbitrarily Scalable Multiprocessor Systems. 4th IEEE Interna-tional Symposium on Electronic Design, Test and Applications(delta 2008), 119-124.Noponen K, Kortelainen K & Seppänen T (2008) Invariant tra-jectory classification of dynamical systems with a case study onECG pattern recognition. Pattern Recognition. In print.Pirttikangas S, Fujinami K & Hosio S (2008) Experiences on datacollection tools for wearable and ubiquitous computing. Int. Sym-posium on Applications and Internet (SAINT2008), Workshopon SensorWebs, Databases and Mining in Networked Sensing Sys-tems (SWDMNSS 2008), IEEE, July 28, Turku, Finland, 149-152.

intelligent systems group (isg) - infotech oulu · 2009-05-18 · oulu institute, and the oulu...

Documents