Download presentation
Presentation is loading. Please wait.
1
Laboratoire d'InfoRmatique en Images et Systèmes d'information UMR 5205 03/06/2015 Vasile-Marian Scuturici and Dejene Ejigu LIRIS-UMR 5205 CNRS, INSA de Lyon Positioning Support in Pervasive Environments Presented at ICPS'06 IEEE International Conference on Pervasive Services 2006 26-29 June 2006, Lyon, France
2
03/06/2015 Topics Background Pervasive computing Positioning needs Related works Background Pervasive computing Positioning needs Related works Model for indoor location detection Learning phase Prediction Phase Model for indoor location detection Learning phase Prediction Phase Experimental results and usage scenario Conclusions and future work
3
03/06/2015 What is pervasive computing? Typical view of a pervasive environment crowded with varieties of ubiquitous devices surrounding a user. A computing trend towards using increasingly ubiquitous and interconnected computing devices in the environment. Enhanced by a convergence of advanced electronic, wireless technologies, and the Internet. Devices involved are very tiny, sometimes invisible, either mobile or embedded in almost any type of object imaginable.
4
03/06/2015 Positioning needs … PSAQL in our platform PerSE to express user intensions: USE sunrise.ppt ON BASE notebook WITH SERVICE projector PSAQL in our platform PerSE to express user intensions: USE sunrise.ppt ON BASE notebook WITH SERVICE projector To ensure neighbourhood constraint, the query can be rewritten as: USE sunrise.ppt ON BASE notebook WITH SERVICE projector IN NEIGHBOURHOOD To ensure neighbourhood constraint, the query can be rewritten as: USE sunrise.ppt ON BASE notebook WITH SERVICE projector IN NEIGHBOURHOOD Here the user assumes that the video projector is situated in the same visually and physically accessible space
5
03/06/2015 Positioning needs … The question is: How can the NEIGHBOURHOOD space be identified in PerSE? The question is: How can the NEIGHBOURHOOD space be identified in PerSE? Neighbourhood relation here is expressed not by the physical proximity like in coordinate positioning system but by perception of the presence in the same bounded space/room.
6
03/06/2015 Related works Among localization and distance measuring methods are: Global Positioning Systems Radio Frequency (RF) delay measurement Association to nearest Access Point Received RF signal strength Among localization and distance measuring methods are: Global Positioning Systems Radio Frequency (RF) delay measurement Association to nearest Access Point Received RF signal strength GPS systems are good for outdoor positioning services The others are based on triangulation or TIX methods and they assume prior knowledge of position of the access point infrastructure GPS systems are good for outdoor positioning services The others are based on triangulation or TIX methods and they assume prior knowledge of position of the access point infrastructure
7
03/06/2015 Topics Background Pervasive computing positioning needs Related works Background Pervasive computing positioning needs Related works Model for indoor location detection Learning phase Prediction Phase Model for indoor location detection Learning phase Prediction Phase Experimental results and u sage scenario Conclusions and future work
8
03/06/2015 Modeling indoor positioning Learning phase: data is collected, classified to create the prediction model Prediction phase: Location prediction based on the real-time data values Learning phase: data is collected, classified to create the prediction model Prediction phase: Location prediction based on the real-time data values Architecture of our learning and prediction model Does not assume prior knowledge of position of APs Based on database methods Does not assume prior knowledge of position of APs Based on database methods (eg. PDA_David locatedIn Common_Room)
9
03/06/2015 Learning phase … Topology of the floors used in our experiments A person holding a PDA moves around the rooms in the building including meeting halls, offices, common rooms, printing rooms and corridors Our WiFi-Spotter and management program is used to track, process and store received signal strength from all n visible access points at each tracking location.
10
03/06/2015 … Learning phase … For each tracking point i in room k, we have a vector with the signal strength values from the APs and a label corresponding to the literal name of the place (room) where the point is situated. Sample attribute-value table showing tracked values.
11
03/06/2015 … Learning phase Signal strength values are classified for pattern identification using data mining tool (MCubiX implementation of the decision tree algorithm). The result from this process is our working model that can later be used for real-time location detection. The model is represented in the predictive model mark-up language – PMML - format.
12
03/06/2015 Prediction phase The two important input parameters for prediction are: Decision rules obtained from the prediction model Real-time signal strength values collected at a specific location The two important input parameters for prediction are: Decision rules obtained from the prediction model Real-time signal strength values collected at a specific location Sample prediction model using two APs and three rooms.
13
03/06/2015 Background Pervasive computing positioning needs Related works Background Pervasive computing positioning needs Related works Background Pervasive computing positioning needs Related works Background Pervasive computing positioning needs Related works Topics Background Pervasive computing positioning needs Related works Background Pervasive computing positioning needs Related works Model for indoor location detection Learning phase Prediction Phase Model for indoor location detection Learning phase Prediction Phase Experimental results and u sage scenario Conclusions and future work
14
03/06/2015 Experimental results … The size of the PMML file containing the model generated after about 4 hours of tracking experiment using three devices is about 320 KB (200rules) and it is within the storage range of mobile devices. Using a cross validation, the results are very encouraging with the error rate below 5%, corresponding to a 95% hit rate. The size of the PMML file containing the model generated after about 4 hours of tracking experiment using three devices is about 320 KB (200rules) and it is within the storage range of mobile devices. Using a cross validation, the results are very encouraging with the error rate below 5%, corresponding to a 95% hit rate. Sample rules generated as a prediction model
15
03/06/2015 … Experimental results Using the Principal Component Analysis (PCA) algorithm, projection of multidimensional data from all visible APIs into 2 dimensions space shows that the data is well separable.
16
03/06/2015 Usage scenario … Consider a scenario where Dave is given a multimedia entertainment service on his PDA while he is in the common room for the tea break. SHARE SERVICE multimedia_player ON BASE LOCALHOST WITH SERVICE multimedia_player ON BASE ALL IN NEIGHBORHOOD The common room is also used by some friends of David. They too are also equipped with PDAs. David wants to share the seen of his video with his friends. In this case he will use the middleware PerSE to express his intention in PSAQL.
17
03/06/2015 … Usage scenario Location prediction combined with context information to determine David’s intension in proactively. IF BASE = LOCALHOST AND BASE_NAME = “PDA_DAVID” AND LOCATION = “CommonRoom” AND RunningAction = “USE * WITH SERVICE multimedia_player ON BASE LOCALHOST” THEN TriggerAction = “SHARE SERVICE multimedia_player ON BASE LOCALHOST WITH SERVICE multimedia_player ON BASE ALL IN NEIGHBORHOOD” The primary role of the prediction model is in this example is to detect that PDA_DAVID is in the Common_Room. It continues detecting who else is present in the room.
18
03/06/2015 Background Pervasive computing positioning needs Related works Background Pervasive computing positioning needs Related works Background Pervasive computing positioning needs Related works Background Pervasive computing positioning needs Related works Topics Background Pervasive computing positioning needs Related works Background Pervasive computing positioning needs Related works Model for indoor location detection Learning phase Prediction Phase Model for indoor location detection Learning phase Prediction Phase Experimental results and usage scenario Conclusions and future work
19
03/06/2015 Future works Integrating this module in to the PerSE middleware Further study on how to avoid hardware influence on the prediction model. Preliminary investigations show that type of signal tracking devices has significant effect on the data. Average values of signals measured by three different devices at the same location (Dell-Axim3 , Dell-Axim50 , HP-Hx4700 )
20
03/06/2015 Conclusions Indoor neighbourhood relation between users is represented: Not by the physical proximity, but by the perception of the presence in the same physically or visually bounded place Indoor neighbourhood relation between users is represented: Not by the physical proximity, but by the perception of the presence in the same physically or visually bounded place We have presented our positioning model for pervasive neighbourhood relationship using the room/office positioning information using database methods The result is found encouraging wit 95% hit rate
21
03/06/2015 Topics Background Pervasive computing positioning needs Related works Model for indoor location detection Learning phase Prediction Phase Experimental results and usage scenario Conclusions and future work Background Pervasive computing positioning needs Related works Model for indoor location detection Learning phase Prediction Phase Experimental results and usage scenario Conclusions and future work Thank You !!
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.