OpenSense           OpenSense	            Karl	  Aberer,	  EPFL	  Boi	  Fal6ngs,	  Alcherio	  Mar6noli,	  	          Mar6n...
OpenSense                          Overview	  •    Research	  challenges	  •    Research	  progress	  and	  results	  •   ...
OpenSense                                Air	  Pollu6on	  •  Air	  pollu6on	  in	  urban	  areas	  is	  a	     global	  co...
OpenSense               Air	  Pollu6on	  Monitoring	  •  Precise	  loca6on-­‐dependent	  and	  real-­‐6me	     informa6on	...
OpenSense                 Opportunity	  •  Monitoring	  today	  	      –  few	  sta6onary	  and	  expensive	  sta6ons	    ...
OpenSense     Value	  of	  Dense	  Measurements	  •  Tradi6onal	  approach	                   •  Recent	  results	      – ...
OpenSense                        Overview	  •  Mo6va6on	  •  Research	  progress	  and	  results	  •  Deployments	  •  Con...
OpenSense                                                           Research	  Challenge	                 SENSING	  SYSTEM...
OpenSense                       Technical	  Challenges	  •  Wireless	  sensing	  devices	       –  energy	  efficiency,	  da...
OpenSense                  What	  is	  the	  problem?	  •  A	  measurement	  system	  such	                 •  Illustra6on...
OpenSense                           U6lity-­‐based	  Control	                                              ApplicaDon	  mo...
OpenSense                                         Testbed	     Sensors	                                                 De...
OpenSense                      Overview	  •  Mo6va6on	  •  Research	  challenges	  •  Deployments	  •  Conclusion	  
OpenSense                        Overview	  •  Mo6va6on	  •  Research	  challenges	  •  Deployments	  •  Conclusion	  
OpenSense                              Sensor	  Behavior	            Open	  sampling	                                     ...
OpenSense                      On-­‐the-­‐Fly	  Calibra6on	  •  Challenge:	  	       –  Supplied	  calibra6on	  may	  not	...
OpenSense             High-­‐resolu6on	  measurement	  Interpola6ng	  measurements	  of	   Planned	  work	  two	  Opensens...
OpenSense                                 Mobility	  Modeling	  Goal	                                                     ...
OpenSense                               Route	  Scheduling	  •  Given	      –  Area	  of	  interest	  Ω	  (Zurich)	      –...
OpenSense                       Air	  Pollu6on	  Models	  •  Forward	  Reasoning	       –  Spa6al	  and	  temporal	  inter...
OpenSense                    A	  Region-­‐Based	  Model	  •  Exis6ng	  grid-­‐based	  models	        –  computa6onally	  e...
OpenSense           Mul6-­‐model	  Query	  Processing	  	  in	              Mobile	  Geosensor	  Networks	  	  •  Approach...
OpenSenseModel-­‐Based	  Query	  Processing	  	      Over	  Uncertain	  Data	  what	  is	  the	  probability	  that	  Bob	...
OpenSenseModel-­‐based	  Anomaly	  Detec6on	                 original	  data	  stream	                              ↓	    ...
OpenSense       Cloud-­‐based	  Time	  Series	  Management	  •  TimeCloud:	  A	  Cloud	  System	  for	  Massive	  Time	   ...
OpenSense                  Sensor	  Context	  Extrac6on	  Objec6ve:	  	  Automa6cally	  annota6ng	  trajectories	  of	  di...
OpenSense     User	  Privacy	  vs.	  Data	  Reliability	  •  Mobile	  devices	  with	  sensing	               •  Privacy	 ...
OpenSense          Privacy	  Protec6on	  Approach	  •  Trust	  authority	  (e.g.	  telco)	     knows	  iden6ty	  and	     ...
OpenSense                        Overview	  •  Mo6va6on	  •  Research	  challenges	  •  Research	  progress	  and	  result...
OpenSense Deployment	  Status	  Basel/Sapaldia	  Sapaldia	  study	                                   Status	  •  Swiss	  T...
OpenSense           Deployment	  Status	  Lausanne	  2	  prototype	  sta6onary	  sta6ons	  and	  1	  prototype	  mobile	  ...
OpenSense                Deployment	  Status	  Zürich	  •  1	  node	  @NABEL	  sta6on	  in	  Dübendorf	  	     (for	  refe...
OpenSenseOzone-­‐Sensor	                  CO-­‐Sensor	                                              Processor	            ...
OpenSense                      Calibra6on	  of	  CO	  Sensor	  @EMPA	  Lab	                                               ...
OpenSense                 Installa6on	  @NABEL	  Dübendorf	  Originally	  calibrated	   O3	  sensor:	  correct	   trend,	 ...
OpenSense             OpenSense	  Visualiza6on	  Portal	            Visualiza6on	  Server	  GSN	                   sensor ...
OpenSense                          OpenSense	  CrowdMap	  The	  data	  from	  NABEL	  sta6ons	  are	  already	  integrated...
OpenSense                          Overview	  •    Mo6va6on	  •    Research	  challenges	  •    Research	  progress	  and	...
OpenSense                               Conclusion	  •  End-­‐to-­‐end	  system	  view	  crucial	      –  Inves6gate	  all...
OpenSense                                                              Team	  •    Karl	  Aberer,	  EPFL-­‐LSIR,	  project...
Upcoming SlideShare
Loading in …5
×

OpenSense

1,167 views

Published on

Published in: Technology, Business
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
1,167
On SlideShare
0
From Embeds
0
Number of Embeds
37
Actions
Shares
0
Downloads
13
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

OpenSense

  1. 1. OpenSense OpenSense   Karl  Aberer,  EPFL  Boi  Fal6ngs,  Alcherio  Mar6noli,     Mar6n  Ve<erli,  EPFL   Lothar  Thiele,  ETH  Zürich  
  2. 2. OpenSense Overview  •  Research  challenges  •  Research  progress  and  results  •  Deployments  •  Conclusion  
  3. 3. OpenSense Air  Pollu6on  •  Air  pollu6on  in  urban  areas  is  a   global  concern   –  affects  quality  of  life  and  health   –  urban  popula6on  is  increasing  •  Air  pollu6on  is  highly  loca6on-­‐ dependent   –  traffic  chokepoints   –  urban  canyons   –  industrial  installa6ons  
  4. 4. OpenSense Air  Pollu6on  Monitoring  •  Precise  loca6on-­‐dependent  and  real-­‐6me   informa6on  on  air  pollu6on  is  needed  •  Officials   –  environmental  engineers:  loca6on  of   pollu6on  sources   –  municipali6es:  crea6ng  incen6ves  to   reduce  environmental  footprint   –  public  health  studies  •  Ci6zens     –  advice  for  outside  ac6vi6es   –  assessment  of  long-­‐term  exposure   –  pollu6on  maps    
  5. 5. OpenSense Opportunity  •  Monitoring  today     –  few  sta6onary  and  expensive  sta6ons   –  models  that  extrapolate  from  pollu6on   sources   Nabel  sta6on  Zürich   –  data  mostly  inaccessible  to  the  public  •  Opportuni6es   mobile  nodes   wireless   fixed  nodes   –  wireless  communica.on:  deploy  larger    Nabel  sta6on  Zürich   numbers  of  sta6ons   –  mobility:  deploy  mobile  sta6ons   –  mobile  devices:  gather  context   informa6on  and  deploy  applica6ons  for   GPRS   ci6zens   GPS  
  6. 6. OpenSense Value  of  Dense  Measurements  •  Tradi6onal  approach   •  Recent  results   –  Few  sta6ons   –  Massive  deployment  of   –  Low  resolu6on  interpolated   sta6ons  (150)  at  street-­‐level   es6mates  of  pollutant   (2008/2009  New  York  City   concentra6ons  across   Community  Air  Quality   massive  regions   Survey)   –  Pollutants  of  interest  heavily   concentrated  along  roads   with  high  traffic  densi6es  
  7. 7. OpenSense Overview  •  Mo6va6on  •  Research  progress  and  results  •  Deployments  •  Conclusion  
  8. 8. OpenSense Research  Challenge   SENSING  SYSTEM   INFORMATION  SYSTEM  NANO   From  many  wireless,  mobile,   …  to  reliable,  understandable  and     TERA   heterogeneous,  unreliable  raw   Web-­‐accessible  real-­‐Dme  informaDon   measurements  …   mobile   wireless   fixed   nodes    Nabel  sta6on  Zürich   nodes   sensor  network  control   opDmizaDon  of  data  acquisiDon  GPRS   informaDon  disseminaDon  GPS   •  More  data,  more  noise,  but  also  more  redundancy   –  Can  we  produce  be<er  quality  data?   •  Exemplary  use  case  for  other  environmental  phenomena   –  Radia6on,  noise,  energy  
  9. 9. OpenSense Technical  Challenges  •  Wireless  sensing  devices   –  energy  efficiency,  data  transmission  and  compression,  sensors  control    •  Mobile  sensors   –  sampling  under  mobility,  data  collec6on  and  dissemina6on  with  mobile   devices,  freshness  of  data,  stream  data  management  •  Community  sensing   –  privacy  protec6on,  trustworthiness  of  data,  relevance  of  data  gathered  and   informa6on  produced    •  Modelling   –  behaviour  and  mobility  of  sensing  devices  è      sensor,  device  and  mobility  models   –  air  quality  informa6on  from  raw  data  è        air  quality  models   –  behaviour,  interests  and  mobility  of  informa6on  consumers  è        privacy,  trust  and  acDvity  models    
  10. 10. OpenSense What  is  the  problem?  •  A  measurement  system  such   •  Illustra6on   as  OpenSense  is  a  complex   1.  Node  decides  individually   system   depending  on  its  state,  e.g.   –  layers   energy   –  dependencies   2.  Nodes  communicate  WSN   and  coordinate   –  dynamicity   3.  Base  sta6on  schedules  nodes  •  Op6miza6on  becomes  a   4.  Mobility  model:  a  third  node   complex  task     arrives,  don’t  measure!   –  mul6ple  op6miza6on   5.  Air  quality  model:  don’t  need   dimensions     measurement!   –  many  system  components  and   6.  Privacy  model:  node  1  should   layers   measure!   –  feedback   7.  Applica6on  model  (e.g.   health  no6fica6on):  no   Two  mobile  nodes:     measurement  needed!   who  should  measure?  
  11. 11. OpenSense U6lity-­‐based  Control   ApplicaDon  model:   Relevance  and  cost   User  acDvity  model:   Mobility  and  user  state   Trust  and  privacy  model:   Reliability  and  security  Control:  translate  high  level     Data:  translate  low  level    u6lity  to  low  level  u6lity   Air  quality  model:   data  to  high  level  informa6on   Sampling  and  correlaDon   Mobility  model:   PredicDon   Wireless  sensor  network:   Local  coordinaDon   Sensors:     Individual  state  
  12. 12. OpenSense Testbed   Sensors   Deployments   •  CO2,  infrared  based     •  Lausanne:  buses   •  CO  electrochemical     •  Zürich:  trams   •  NO2  electrochemical     •  SO2  electrochemical     •  Basel:  sta6onary  wireless   •  O3  silicon  based     network   •  Fine  par6cles  mechanical   Power  suppliers   pDr1000:  ultrafine  par6cles  Sensorscope   (FH  Nordwestschweiz)  DataLogger     SHT75:  air  temp  and  humidity   Telaire  T6613:  C02   Langan  T15n:  CO   Sensorscope   Smart   Interfaces  
  13. 13. OpenSense Overview  •  Mo6va6on  •  Research  challenges  •  Deployments  •  Conclusion  
  14. 14. OpenSense Overview  •  Mo6va6on  •  Research  challenges  •  Deployments  •  Conclusion  
  15. 15. OpenSense Sensor  Behavior   Open  sampling   Closed  sampling      Sensors  directly  exposed  to   Sensors  exposed  to  measurand  inside  environmental  measurand   controlled  chamber  Benefits:   Benefits:  •  simple  &   slim  solu6on   •  absolute  measurements  •  con6nuous  sampling   •  noise  due  to  environment  filtered  Drawbacks:   Drawbacks:  •  no  absolute  concentra6on  values   •  complex  &  bulky  •  noisy  signal   •  non-­‐con6nuous  sampling  Typical  response:   Typical  response:   IDEA:  Combine  the  two  approaches  and  get  the  benefits  of  both.  
  16. 16. OpenSense On-­‐the-­‐Fly  Calibra6on  •  Challenge:     –  Supplied  calibra6on  may  not  match  project  requirements   –  Baseline  driq  due  to  sensor  aging  •  Approach:   –  Ini6al  calibra6on  using  sta6onary,  high  quality  instruments   –  When  deployed  periodic  recalibra6on  using  mobile  sensor  nodes   Original  calibra6on   performs  with  an   Aqer  recalibra6on   average  error  of   the  average  error   30ppb   drops  below  3ppb  
  17. 17. OpenSense High-­‐resolu6on  measurement  Interpola6ng  measurements  of   Planned  work  two  Opensense  sta6onary   •  Measurements  obtained  along  sta6ons   the  road  network  +  anisotropic  •  A  difference  of  10m  from  road  is   diffusion  on  lines,  tuned  by  traffic   and  popula6on  density     considerable   (from  mobile  sensors)   Sta6on  1058   Sta6on  1059  
  18. 18. OpenSense Mobility  Modeling  Goal   Appropriate  tool:  microscopic  traffic  •  Simulate  realisDc  trajectories  of   simulators  (SUMO,  AIMSUN)   vehicles    Tes6ng  different  control  strategies  before  deployment  •  What  is  the  marginal  benefit  of   adding  an  addi6onal  vehicle/line  to   the  system  •  Knowing  the  traffic  pa<erns,  is  the   system  coverage   suitable  for   regions  with  fluctua6ng  traffic   (emissions)?  •  What  is  the  effect  of  a  traffic  event   on  the  coverage  of  the  system?   3D  view  of  traffic  simulaDon  run  in  front  of  Lausanne  Train   StaDon,  using  SimLo  model  (LAVOC,  EPFL)  
  19. 19. OpenSense Route  Scheduling  •  Given   –  Area  of  interest  Ω  (Zurich)   –  N  measurement  instruments   •  Each  has  a  limited  budget  E     –  M  tram  and  bus  tracks  •  Ques6ons   –  Which  subset  of  tracks  (and  trams)   gives  the  best  coverage  of  the  city?   –  Which  tram  should  measure  over     shared  track  pieces?   •  The  program  is  NP-­‐Complete  
  20. 20. OpenSense Air  Pollu6on  Models  •  Forward  Reasoning   –  Spa6al  and  temporal  interpola6on  of  pollu6on  levels   –  Advanced  warning  for  dangerous  levels  •  Backward  Reasoning   –  Crea6ng  an  emission  inventory   –  Iden6fying  previously  unknown  sources  •  Meta-­‐Reasoning   –  Op6mal  sensor  placement   –  Sparse  sampling  
  21. 21. OpenSense A  Region-­‐Based  Model  •  Exis6ng  grid-­‐based  models   –  computa6onally  expensive  for  fine   grids   –  do  not  dis6nguish  streets  •  Pollu6on  dispersion  is  not  uniform   within  a  grid   –  Ground-­‐level  air  pollu6on  is  heavily   influenced  by  streetscape  and  land  use   –  A  region-­‐based  model  may  be  more   appropriate  for  OpenSense   ADMS-­‐Urban,  London  2010  
  22. 22. OpenSense Mul6-­‐model  Query  Processing    in   Mobile  Geosensor  Networks    •  Approach   Con$nuous  Moving  Queries   –  Middle  layer  produces  a  model   Give  a  (in  car)  pollu6on  update   cover  from  a  set  of  regression   Aggregate  Queries   every  30  mins   models  on  an  area   COX  emi<ed  yesterday  in   –  Con6nuous  sensor  updates   Lausanne  center   –  Con6nuous  and  ad-­‐hoc  queries    •  Advantages   Model-­‐based  middle  layer     –  Handling  spurious  updates  to  the   data  base   –  Minimizes  data  storage     –  Query  results  useful  from   DBMS   applica6on  perspec6ve   (storage  of  raw  sensor  values)     Mobile  Sensor  Data     Mobile  Sensor  Data     (Pollu.on  Values)   (Pollu.on  Values)  
  23. 23. OpenSenseModel-­‐Based  Query  Processing     Over  Uncertain  Data  what  is  the  probability  that  Bob  is  at  room  4  at  $me  1?   original  data  stream   ↓   inference  of  Dme-­‐varying  probability  distribuDon   (dynamic  density  metrics)   ↓   creaDng  probabilisDc  views   (Ω-­‐View  builder)  
  24. 24. OpenSenseModel-­‐based  Anomaly  Detec6on   original  data  stream   ↓   approximaDon  using  user-­‐selected  models   ↓   detecDng  anomalies   ↓   user  confirmaDon:  anomaly  is  an  actual  error?  
  25. 25. OpenSense Cloud-­‐based  Time  Series  Management  •  TimeCloud:  A  Cloud  System  for  Massive  Time   Series  Management  •  Key  features   –  manages  large-­‐scale  6me  series  in  the  cloud     –  scalable,  fault-­‐tolerant   –  built  upon  Hadoop  and  Hbase   –  adap6ve  data  storage  through  par66on-­‐and-­‐ cluster   –  model-­‐based  cache  for  fast  model-­‐based   views     –  model-­‐coding  join  for  fast  distributed  join   based  on  bitmap  representa6on  of  6me   series.      
  26. 26. OpenSense Sensor  Context  Extrac6on  Objec6ve:    Automa6cally  annota6ng  trajectories  of  different  types  of  moving  objects  (cars,  people)   bus   metro   walking   Seman$c   trajectory   home    office   market   home   Seman$c  Annota$on  Middleware   Hidden Spatial Map Markov Join Matching Model region   road  network   point  of  interest   e1   e2   e3   e4   e5   e6   e7   GPS   episodes  
  27. 27. OpenSense User  Privacy  vs.  Data  Reliability  •  Mobile  devices  with  sensing   •  Privacy  protec6on  mechanisms  try   capabili6es   to  break  the  link  between  data  and   its  source   –  ParDcipatory  sensing   –  E.g.  environmental  sensing,   •  Thus,  there  is  a  clear  trade-­‐off   health-­‐care  monitoring,  etc.   between  privacy  and  •  Incen6ves  for  par6cipa6on   trustworthiness  of  data  sources   –  Privacy  concerns   •  Iden6ty   •  Loca6on   –  Trustworthiness   Sensor,  air  polluDon,  mobility,  behavior   models  used  to  esDmate  reliability  of  data  
  28. 28. OpenSense Privacy  Protec6on  Approach  •  Trust  authority  (e.g.  telco)   knows  iden6ty  and   Aggregation trustworthiness  of  users   Server•  Aggrega6on  server  receives   Trust trust-­‐rated  but  privacy-­‐ Scores Ratings preserving  data   –  Anonymize  data  sources   –  Obfuscate  data,  loca6on-­‐  or   6me-­‐stamps   Trust –  Hide/add  events   Authority Honest  and  malicious     Entropy  as  measure  for   measurements     uncertainty  about  user   clearly  dis6nguished   data  remains  high  
  29. 29. OpenSense Overview  •  Mo6va6on  •  Research  challenges  •  Research  progress  and  results  •  Conclusion  
  30. 30. OpenSense Deployment  Status  Basel/Sapaldia  Sapaldia  study   Status  •  Swiss  Tropical  and  Public  Health   •  Calibra6on  tests  performed  in   Ins6tute  of  Basel  University   2010  •  Es6mate  individual  exposure   •  Sta6onary  sta6ons  will  be   indoors  and  outdoors   delivered  on  May  18  Sapaldia  will  use  sta6ons  for    indoor  air  quality  monitoring  
  31. 31. OpenSense Deployment  Status  Lausanne  2  prototype  sta6onary  sta6ons  and  1  prototype  mobile  sta6on    •  Currently  under  tes6ng  at  EPFL  •  Mobile  sta6on  will  be  mounted  on  a   bus  on  May  23  Measured  parameters  •  NO2,  CO  (2  sensors),  Humidity,   Temperature,     CO2  (only  mobile  sta6on)  Power  •  Solar  panel  (sta6onary  sta6ons)  •  Bus  power  (mobile  sta6on)  Data  •  Transmission  via  GPRS  to  a  central   server   Sta6on  1058   Sta6on  1059  
  32. 32. OpenSense Deployment  Status  Zürich  •  1  node  @NABEL  sta6on  in  Dübendorf     (for  reference  measurements):   –  Communica6on:  GSM,  WLAN   –  Sensors:  2  x  O3,  CO,  temperature/humidity   –  GPS  •  1  node  on  top  of  Tram              in  Zürich  is  in  prepara6on   14   (mid.  July  2011):   –  Communica6on:  GSM,  WLAN   –  Sensors:  O3,  temperature/humidity   –  GPS   –  Accelerometer  •  2  further  nodes  in  construc6on  (July)  
  33. 33. OpenSenseOzone-­‐Sensor   CO-­‐Sensor   Processor   WLAN   GPS   GSM   USB-­‐Hub  
  34. 34. OpenSense Calibra6on  of  CO  Sensor  @EMPA  Lab   gas  bo<le   empty  Ini6ally  not   calibrated   calibrated  
  35. 35. OpenSense Installa6on  @NABEL  Dübendorf  Originally  calibrated   O3  sensor:  correct   trend,  but  wrong   absolute  value.  Calibra6on  required.  
  36. 36. OpenSense OpenSense  Visualiza6on  Portal   Visualiza6on  Server  GSN   sensor data   cache   Significant     Change  Condi6on   Interpola6on   Image  grid   cache  
  37. 37. OpenSense OpenSense  CrowdMap  The  data  from  NABEL  sta6ons  are  already  integrated.  It  is  possible  to  add  data  via  SMS,  Email  or  online   Form.  Based  on  open  source  plaworm.     OpenSense  CrowdMap  is  not  yet  publicly  available.  
  38. 38. OpenSense Overview  •  Mo6va6on  •  Research  challenges  •  Research  progress  and  results  •  Deployments  
  39. 39. OpenSense Conclusion  •  End-­‐to-­‐end  system  view  crucial   –  Inves6gate  all  system  layers:  sensor  –  user  interfaces   –  U6lity-­‐based  framework  as  integra6ve  approach  •  Results  applicable  beyond  air  pollu6on   –  Complex,  distributed,  par6cipatory  measurement  •  Involvement  of  Nokia   –  Personalized  health  applica6on  •  For  more  informa6on:  opensense.epfl.ch  
  40. 40. OpenSense Team  •  Karl  Aberer,  EPFL-­‐LSIR,  project  leader           •  Alcherio  Mar6noli,  EPFL-­‐DISAL,  PI     –  Thanasis  Papaioannou,  postdoc     –  Chris  Evans,  PhD     –  Dipanjan  Chakraborty,  (on  leave  from   –  Emanuel  Droz,  engineer     IBM  Research  India),  visi6ng  researcher     –  Adrian  Arfire,  PhD     –  Hoyoung  Jeung,  postdoc     •  Lothar  Thiele,  ETH  Zürich,  PI     –  Rammohan  Narendula,  PhD     –  Olga  Saukh,  postdoc     –  Mehdi  Riahi,  PhD     –  Jan  Beutel,  postdoc     –  Zhixian  Yan,  PhD     –  Jayashree  Ajay-­‐Candadai,  PhD     –  Sofiane  Sarni,  engineer     –  Alex  Arion,  PhD     –  Saket  Sathe,  PhD    •  Mar6n  Rajman,  EPFL-­‐LIA,  coordinator    •  Boi  Fal6ngs,  EPFL-­‐LIA,  PI     –  Jason  Jingshi  Li,  postdoc    •  Mar6n  Ve<erli,  EPFL-­‐LCAV,  PI     –  Guillermo  Barrenetxea,  postdoc     –  Andrea  Ridolfi,  postdoc     –  Heather  Miller,  PhD    

×