View
215
Download
0
Tags:
Embed Size (px)
Citation preview
Learning the meaning of Learning the meaning of placesplaces
IfGiIfGi
Location based ServicesLocation based Services
SS 06SS 06
Milad SabersamandariMilad Sabersamandari
InhaltInhalt
IntroductionIntroduction Existing place learning algorithmsExisting place learning algorithms Extracting Places from traces of Extracting Places from traces of
locationslocations Application with BluetoothApplication with Bluetooth Advantages and disadvantagesAdvantages and disadvantages ReferencesReferences
IntroductionIntroduction
Location learning systemsLocation learning systems Locations are expressed in 2 Locations are expressed in 2
principal waysprincipal ways• CoordinatesCoordinates• LandmarksLandmarks
Intrested in „places“ (e.g. home, Intrested in „places“ (e.g. home, work, cinema)work, cinema)
IntroductionIntroduction
Define „places“Define „places“ Manually by handManually by hand
• Rectangular region around an office Rectangular region around an office represented in coordinatesrepresented in coordinates
AutomaticallyAutomatically• Spends a significant amout of time Spends a significant amout of time
or/and visits frequentlyor/and visits frequently -> Place learning algorithms-> Place learning algorithms
IntroductionIntroduction
Locations based servicesLocations based services• Location based reminderLocation based reminder• Location based to-do list applicationLocation based to-do list application• „„Location based intelligent desicions Location based intelligent desicions
service“service“
Existing place learning algorithmsExisting place learning algorithms
Ashbrook and Starner´s GPS Dropout Ashbrook and Starner´s GPS Dropout Hierachical Clustering Algorithm Hierachical Clustering Algorithm (A&S)(A&S)
The comMotion Recurring GPS The comMotion Recurring GPS Dropout AlgorithmDropout Algorithm
The BeaconPrint AlgorithmThe BeaconPrint Algorithm
Ashbrook and Starner´s Clustering Ashbrook and Starner´s Clustering Algorithm (A&S)Algorithm (A&S)
Loss of GPS signal of at least Loss of GPS signal of at least tt minutesminutes
Indicates a speed of continuilly below Indicates a speed of continuilly below 1 mile per hour1 mile per hour
Positions are merged (variant k-Positions are merged (variant k-means clustering algorithm)means clustering algorithm)
The comMotion Recurring GPS The comMotion Recurring GPS Dropout AlgorithmDropout Algorithm
GPS is lost three or more times GPS is lost three or more times within a given radiuswithin a given radius
Merge the points to placesMerge the points to places
The BeaconPrint AlgorithmThe BeaconPrint Algorithm
Fingerprint algorithmFingerprint algorithm Input: sensor log from mobile deviceInput: sensor log from mobile device List of places the device went List of places the device went
(waypointlist)(waypointlist) GSM and 802.11GSM and 802.11
The BeaconPrint AlgorithmThe BeaconPrint Algorithm
1. Segment a sensor log into times when 1. Segment a sensor log into times when the device was in a stable place and the device was in a stable place and assign a waypoint.assign a waypoint.
2. Merge waypoints which are captured 2. Merge waypoints which are captured from repeat visits to the same place.from repeat visits to the same place.
Likewise, an effective recognition Likewise, an effective recognition algorithm has two capabilities:algorithm has two capabilities:• 1. Recognize when the device returns to a 1. Recognize when the device returns to a
known place using a waypoint list.known place using a waypoint list.• 2. Recognize when the device is not in a place 2. Recognize when the device is not in a place
We refer to this state as mobile.We refer to this state as mobile.
Extracting Places from traces of Extracting Places from traces of locationslocations
Uses Place Lab to collect traces of Uses Place Lab to collect traces of locationslocations
In many cities and towns available In many cities and towns available Place Lab works in urban areas Place Lab works in urban areas
aswell as indoorsaswell as indoors Location recorded once per secondLocation recorded once per second Places appear as clusters of locationsPlaces appear as clusters of locations
Extracting Places from traces of Extracting Places from traces of locationslocations
Place LabPlace Lab• Uses that each WiFi access point Uses that each WiFi access point
broadcasts its unique MAC addressbroadcasts its unique MAC address• A database maps these addresses to A database maps these addresses to
longitude and latidute coordinates longitude and latidute coordinates
Existing clustering AlgorithmExisting clustering Algorithm
k-means Algorithmk-means Algorithm Gaussian mixture model (GMM)Gaussian mixture model (GMM) Require the number of clusters as a Require the number of clusters as a
parameterparameter Require a significant amout of Require a significant amout of
computationcomputation
Time based clusteringTime based clustering
Eliminate the intermediate locations Eliminate the intermediate locations between important placesbetween important places
Determine the number of clusters Determine the number of clusters (important places) autonomously(important places) autonomously
Simple enough to run on a simple low Simple enough to run on a simple low battery mobile devicebattery mobile device
Time based clusteringTime based clustering
Basic idea is to cluster along the time Basic idea is to cluster along the time axisaxis
New measured location is compared New measured location is compared with previous locationswith previous locations
Decide if the mobile device is movingDecide if the mobile device is moving Parameter:distance Parameter:distance d d between the between the
locations and a cluster´s time locations and a cluster´s time duration duration tt
Time based clusteringTime based clustering
Parameter: distance Parameter: distance dd, time , time tt Current cluster Current cluster clcl Pending location Pending location plocploc Significant places Significant places PlacesPlaces
Time based clusteringTime based clustering
Unlike other clustering algorithms Unlike other clustering algorithms this algorithm computes the clusters this algorithm computes the clusters incrementallyincrementally
The computation is simpleThe computation is simple Easily supported on small battery Easily supported on small battery
mobile devicesmobile devices
Application with BluetoothApplication with Bluetooth
Bluetoothcell with radius Bluetoothcell with radius rr Bool value for each cellBool value for each cell Short distanceShort distance Time duration of 11 secondsTime duration of 11 seconds
Application with BluetoothApplication with Bluetooth
ReplaceReplace• Measured location Measured location loc loc
measured BTcell measured BTcell cellcell• Pending locationPending location ploc ploc
pending BTcellpending BTcell pcell pcell Current clusterCurrent cluster cl cl
as a set of BTcellsas a set of BTcells
Advantages and disadvantagesAdvantages and disadvantages
GPS (Advantages)GPS (Advantages)• Standardized Standardized • Covers most of the earth´s surfaceCovers most of the earth´s surface• Continually decreasing in costContinually decreasing in cost
GPS (Disadvantages)GPS (Disadvantages)• Inability to function indoorsInability to function indoors• Occasional lack of geometry accuracyOccasional lack of geometry accuracy• Loss of signal in urban canyons and Loss of signal in urban canyons and
other „shadowed“ areasother „shadowed“ areas
Advantages and disadvantagesAdvantages and disadvantages
Bluetooth (Advantages)Bluetooth (Advantages)• Standardized Standardized • 3 classes (different ranges)3 classes (different ranges)• Everywhere available (indoor) Everywhere available (indoor)
Bluetooth (Disadvantages)Bluetooth (Disadvantages)• Short distanceShort distance• Long time durationLong time duration• Accuracy = 1 BluetoothcellAccuracy = 1 Bluetoothcell• Bad java supportBad java support
ReferencesReferences1.1. Jong Hee Kang, William Webourne, Benjamin Stewart, Gaetano Jong Hee Kang, William Webourne, Benjamin Stewart, Gaetano
Borrielo. Borrielo. Extracting Places from Traces of LocationsExtracting Places from Traces of Locations2.2. Jeffrey Hightower, Sunny Consolvo, Anthony LaMarca, Ian Smith, Jeffrey Hightower, Sunny Consolvo, Anthony LaMarca, Ian Smith,
Jeff Hughes†. Jeff Hughes†. Learning and Recognizing the Places We GoLearning and Recognizing the Places We Go3.3. John Krumm, Ken Hinckley. John Krumm, Ken Hinckley. The NearMe Wireless Proximity The NearMe Wireless Proximity
ServerServer