130 likes | 203 Views
Explore Context Fusion method using Dynamic Bayesian Networks and Fuzzy Logic for accurate inference from sensor data with focus on source reliability. Evaluate system performance in location estimation using WLAN Access Points and Infrared Beacons.
E N D
Context Fusion: Dealing with Sensor Reliability Christos Anagnostopoulos Odysseas Sekkas Stathes Hadjiefthymiades Pervasive Computing Research Group, http://p-comp.di.uoa.gr Department of Informatics and Telecommunications University of Athens, Greece SensorFusion07, 08.10.2007, Pisa, Italy
Context Fusion • Context Estimation is characterized by imprecise knowledge (e.g. missing information and unreliability of sources) • Context Fusion is the method of deriving high-level context from low-level, inaccurate sensor data. • Context Fusion Engine • Dynamic Bayesian Networks (DBN) and Fuzzy Logic • incorporates the reliability of sources • more accurate inference on the current user situation i.e., a set of aggregated pieces of context.
Reliability of Sources The context-determination rule that concludes a situationp w.r.t, reliability of the sources: [(a1is u1) andconf1] … [(anis un) andconfn] (pis u) confidence value confidence value ai= contextual ingredient (attribute) ui = value confi= confidence of sensor readings on measuring ui
Probabilistic Fusion • Random variables of the DBN are • attributes a (i.e., sensor readings) • situation p (i.e. location of the user, actions, etc.)
Probabilistic Fusion The calculation of conditional probabilities determines the value of the situation at time t i.e., p = p(t) Fusion: find the situation p(t) that maximizes P(p(t))
Fuzzy Probabilistic Fusion Fuzzy Probabilistic Fusion result v* Fuzzy Inference confp, v Confidence Probabilistic Fusion p situation Determination Rule … a2 a1 aN … Probabilistic Fusion results conf1, v*1 conf2, v*2 confN, v*N
Fuzzy Sets for Confidence Fuzzy Values for P(p(t)), confp and confidence probability P*(p(t)) denoted as Linguistic Terms.
Fuzzy Inference Rules ifP(p(t))is lowthenP*(p(t)) is low ifP(p(t))is mediumandconfpis low thenP*(p(t)) is very low ifP(p(t))is mediumandconfpis high thenP*(p(t)) is somewhat high ifP(p(t))is highandconfpis lowthenP*(p(t)) is medium ifP(p(t))is highandconfpis highthenP*(p(t)) is high
System Evaluation We assume that situation p(t) is the location L of the user at time t,L = {meeting room, entrance,…}
System Evaluation • Test-bed involving two technologies • WLANAccess Points (4) • Infrared Beacons (5) • Test-bed area: UoA, Dept. of Informatics and Telecommunications
Reliability of Sources Probability distribution for the sensor AP1 P(AP1=v1|L=L1)=0.5 Reliability (h) for each sensor (A=Access Point (AP), B = IR Beacon) • IR-Beacons appear more reliable on location estimation than WLAN APs • IR-Beacons have shorter range of emission thus improving the accuracy of the estimated location
Thank you http://p-comp.di.uoa.gr