Keynote Speaker - Todd Humphreys: Low-Cost Centimer-Accurate Mobile Positioning

Todd Humphreys

Transcript

1. Low-Cost Centimeter-Accurate Mobile Positioning Todd Humphreys, Ken Pesyna, Daniel Shepard Department of Aerospace Engineering and Engineering Mechanics; Radiosense LLC University of Texas at Austin Texas  GIS  Forum  |  Oct.  29,  2015
2. “… In that Empire, the Art of Cartography attained such Perfection that the map of a single Province occupied the entirety of a City, and the map of the Empire, the entirety of a Province. In time, those Unconscionable Maps no longer satisfied, and the Cartographer’s Guilds struck a Map of the Empire whose size was that of the Empire, and which coincided point for point with it.” —Suarez Miranda,Viajes de varones prudentes, Libro IV,Cap. XLV, Lerida, 1658 On Exactitude in Science Jorge Luis Borges, Collected Fictions, translated by Andrew Hurley
3. A  1-­‐to-­‐1  Map  of  the  World   What  would  it  take  to  create  a   global  3D  map  with  geo-­‐referenced     sub-­‐decimeter-­‐accurate  coordinates?  
4. Location Sharing Fitness Tracking Driving Directions Location-based Searches Mass-Market GNSS Positioning Today
5. Where It’s Going Autonomous Driving Large-scale Mapping Virtual Reality Augmented Reality
6. Improvements to SPS GPS Accuracy have Stalled 50   0.8   0.5   2   1   0   2   4   6   2000   2001   2002   2003   2004   2005   2006   2007   2008   2009   2010   2011   2012    Range  Error,  meters   Year   Orbit/Clock   Thermal   MulGpath   Atmospheric   •  Prior  to  2000     –  IntenGonal  degradaGon   –  50+  meter  errors   •  May  2,  2000   –  DegradaGon  turned  off   •  Since  2000   –  Steady  Improvements   –  Leveling  off   –  Meter-­‐level  precision   Source: NOAA National Geodetic Survey Meters Error Average SPS GPS Positioning Error
8. Carrier-Phase-Based GNSS Positioning Centimeter Accurate
9. Mass-Market Demands: Short TTF and Low Cost Two  criGcal  performance  metrics  for  mainstream  cm-­‐accurate   GNSS  will  be  !me  to  fix  and  cost.    Keeping  these  tolerably  low   will  require  network  RTK  or  PPP-­‐RTK  with  a  dense  network:   1.  As  compared  to  tradiGonal  PPP  (sparse  reference  network),   network  RTK  and  PPP-­‐RTK  have  faster  convergence  Gmes   2.  For  both  dual-­‐  and  single-­‐frequency  rover  receivers,  a   Gghter  network  significantly  improves  TTF  down  to  an  inter-­‐ staGon  distance  of  about  15  km   3.  When  targeGng  the  mass  market,  it  makes  economic  sense   to  densify  network:    increased  density  speeds  TTF  and   lowers  receiver  costs  for  millions  of  users      
11. Mass-Market Demands: Short TTF and Low Cost •  Is  carrier-­‐phase-­‐based  GNSS  posiGoning  feasible   on  consumer-­‐grade  mobile  pla`orms  such  as   smartphones  and  tablets?   •  What  are  the  primary  challenges?   •  How  can  these  be  addressed?  
12. December 2014: First successful RTK positioning solution with a smartphone
13. Handheld RTK result with some signals passing through Ken’s body
14. One  of  Picasso’s  famous  light  painGngs      
15. GNSS “light painting” with a smartphone
17. Standard Deviation: 3.4 mm Standard Deviation: 11.4 mm Phase Residuals Survey-Grade Antenna 8.6 mm Phase Residuals Smartphone Antenna Primary Challenge: Large Time-Correlated Multipath Errors
18. •  ExisGng  techniques  include:   –  MulGpath-­‐EsGmaGng  Delay-­‐Lock  Loop  [NeeSie&94],  [TowFen&95]   –  Coupled  mulGpath-­‐esGmaGng  delay-­‐lock,  phase-­‐lock  loop  [PsiErt15]   –  Signal-­‐to-­‐noise-­‐raGo-­‐based  mulGpath  error  correcGon  [AxeCom&96]   –  Enhanced  Strobe  Correlator  [GarRou97]   –  Ray  tracing  [LauCro07]   •  Downsides  are  that  these  exisGng  techniques  require  either:   –  a-­‐priori  knowledge  of  antenna  moGon  profile  [PsiErt15]  or  range  to  nearby  reflecGon  surfaces   [LauCro07]   –  Extra  computaGonal  power  to  generate  measurements  at  more  than  3  correlaGon  taps   [NeeSie&94],  [TowFen&95]   –  A  lengthy  measurement  duraGon  to  idenGfy  mulGpath  frequency  [AxeCom&96]   –  A  high  sampling  rate  in  excess  of  20  megasamples  per  second  [GarRou97]   Existing Multipath Mitigation Techniques Inapt
19. Expensive  to  reduce  the  magnitude  of  mulGpath  errors,  so  instead,   we’ll  reduce  the  correlaGon  Gme  of  mulGpath  via  antenna  moGon   and  appropriate  modeling:                 Phase Residuals (No Motion) Phase Residuals (Motion)
20. EsGmator   Single  ReflecGon   MulGpath  Model   , , (quality) (dynamics) (error statistics) 1.  Position Estimate 2.  Integer Ambiguity Estimates Phase Measurements Modeling Antenna Dynamics and Phase Errors
21. Solve  this  cost  funcGon  for  the  real-­‐  and  integer-­‐ valued  state  elements:  
22. How Much Does Motion Help? (simulation) Smartphone-grade antenna: Motion reduces TTF by >4x
23. How Much Does Motion Help? (simulation) Survey-grade antenna: Motion increases TTF
24. How Much Does Motion Help? (simulation) Survey-grade antenna: Motion increases TTF
25. How Much Does Motion Help? (real data) Low-quality patch antenna: Motion reduces TTF by ~2x
26. How Much Does Motion Help? (real data) Even faster TTF if motion profile can be constrained
27. But our smartphone also has a camera ... ... what could we do with the camera?
28. Over  the  past  15  years,  the   computer  vision  community  has   made  stunning  advances  in  what   is  variously  known  as  structure   from  moGon  (SFM),   photogrammetry,  Vision-­‐based   Simultaneous  LocalizaGon  and   mapping  (VSLAM)  [HarZis00],   [TriMcL&00],  [StraMon&12],   [NueWei&11],  [KleMur07]       OpenMVG, 2015 Agarwal, Snavely, 2009 Colosseum
29. Extending the SFM technique, we can optimally fuse camera images and GNSS phase measurements to shorten TTF and create geo-referenced 3D maps
34. Sparse 3D Reconstruction
35. Dense 3D Reconstruction
36. Jointly  fuse  GNSS  carrier  phase  and  vision  measurement  in  the  same  non-­‐ linear  esGmator:     OpGmal  ML  soluGon   Phase  Measurements   Camera  OrientaGons   Camera  PosiGons   Vision  Measurements   Point  Feature  PosiGons   CDGNSS  Integer  AmbiguiGes   2 2 Our technique preserves both the sparsity of bundle adjustment and the integerness of the carrier phase ambiguities Resulting 3D point cloud is georeferenced (e.g., to ITRF)
38. Result: A globally-referenced 3D point cloud accurate to better than 1 cm