Federating Spatial Coordinate Systems

1
Federating SpatialCoordinate Systems

• Lewis Girod
• Dgroup Presentation
• 9 February 2000

2
Teaser Example Earthquake Disaster Relief

• Partially collapsed building
• GPS antennas on outside of building connect to
  internal nodes
• Acoustic ranging used indoors
• Seismic sensors in floor/walls
• Cameras/mics normally used for security/conferenci
  ng apps
• Task
• Localize, report seismic sources.
• Return a/v of seismic sources.
• Report damage to net building

3
How localization comes into play

• Rescuers need to use GPS to locate the building
  on their maps
• When they get there, they will need fine-grained
  locations of survivors within the building
• In order to photograph possible survivors the
  system must determine precisely where the cameras
  point and which ones might point at a seismic
  event (i.e. a possible survivor)

4
In this example, localization helps in two
general areas

• At the application layer
• Camera needs to learn its position and field of
  view relative to seismic measurements
• Reports must be relative to physical structure of
  building, e.g. floorplan
• At the network layer
• Queries about specific room (e.g. requesting new
  photos) can be directed more efficiently
• Lapses in coverage can be identified

5
Claim Spatial Coordinate Systems are
Fundamentally Useful

• How so? In a nutshell
• Devices take up physical space
• Thus sufficiently fine-grained spatial
  coordinates are a unique attribute with implicit
  routing information
• Location is relevant to many applications
• These devices are doing things in the world
  users need to find them inputs and outputs to
  tasks often reference locations

6
It enables interesting apps

• Benefits at the application layer
• Self-configuring applications
• Deployment Ex. Light switches appropriately
  mapped to switched outlets
• Maintenance Ex. Reconfigure, swap in equivalent
  nearby sensors to maintain tasked coverage as
  system degrades
• Location-infused applications
• Train camera on source of seismic disturbance
• User interfaces
• UI references physical space to control the
  system
• Often topological coordinates useful (3rd truck
  from left)

7
It improves network scaling

• Benefits at the network layer
• Naming uniquely names data, regions, endpoints
• Names are self-configuring and directly relate to
  apps
• Propagation implies a heading to destination
• More efficient diffusion
• As the crow flies can mislead - but it comes
  for free
• Power distance traveled bounds min power cost
• Basis for assessment of network efficiency
• Each hop fixed cost radio energy cost
  Balakrishnan99

8
How can we achieve fine-grained localization?

• Need sensors that can measure distance
• Relative or Global?
• Relative spatial measurements tend to be more
  accurate because the observed phenomena are
  local, shorter ranges, etc.
• Global measurements (e.g. GPS) are much coarser
  (40m) but provide a single coordinate system that
  can be exported unambiguously
• Really need both of these...

9
High Level Architecture

• Combine global scope of GPS with precision of
  relative sensors
• Need ways of fusing local global coordinate
  frames
• Local algorithms, multiple modalities fused into
  local frames
• Coordinates specified at multiple granularities
• Possible caveats
• Cost of coordination do it on-demand?
• Cost of sensing do it all at once?
• Local coordinates live as long as local network
• Anonymity depends on coordination protocol and
  sensors compare GPS which has a self-contained
  passive receiver

10
Relative Measurement Techniques

• Ultrasound time of flight (Active Bat)
• Location defined by range to three points
• Constrained by partition boundaries
• Subject to interference from obstructions
• Low power, high update rate, sub-cm range
  accuracy
• Anonymous local ranging
• Differential GPS
• Cartesian position/orientation relative to
  basestation
• Only works outdoors, but can be quite accurate
  (2cm)
• Large form factor, large lag time and startup
  power cost per isolated measurement
• Anonymous local measurements

11
Relative Measurement Techniques

• Multiple-baseline stereo imaging
• Correlate location of IR emitters in multiple
  views of scene
• Multiple cameras share images and compute tag
  positions
• Low update rate single measurement costly in
  power
• Not as accurate, but may give rough 3-D terrain
  model
• RF time of flight (Pinpoint)
• Accurate and works indoors
• System detects position of low power tags
• Requires installed infrastructure which has a
  large form factor and is not designed for
  low-power operation.

I dont know whether this is fundamental
limitation of their technology...
12
Fusing Relative Coordinates

• Hard part fusing these measurements
• Each modality has different strengths/weaknesses
• GPS costly in power, large form factor, outdoors
  only
• Ultrasound suffers in obstructed environments
• Cameras have low ranging precision, but see
  obstacles
• Fusion is result of localized algorithms
• Sensor-specific merging of adjacent coordinate
  frames
• Use alternate modalities to improve consistency
• Export local coordinates referenced globally
• Layered coordinates Coarse global fix from GPS
  plus fine-grained localization within local
  coordinate frame

13
Revisiting the building example

• Within building, acoustic ranging enables
  fine-grained coord frames, establishes presence
  of partitions
• DGPS antennas on building skin provide accurate
  reference points (helps correct acoustic errors)
  and coarse global coordinates
• Cameras determine locations of sensors in their
  fields of view, coordinate to position them in 3D

14
Where to start?

• Acoustic ranging measure TOF of sound
• Accurate (sub-cm) repeatable (low variance)
• Cheap (easy timing requirements)
• Power efficient (no startup cost, low per-sample
  cost)
• Detects structure of environment (partitions)
• But
• Large errors in obstructed environments
• Limited range means local coordinate frames must
  be fused

Radio pulse requests chirp
?T
Sonic chirp slowly travels back
15
Acoustic ranging example
Coordinate System ABC Coordinate System DEF
B
A
D
E
C
F
Acoustic ranging transceiver Acoustic ranging
transmitter
16
Coordinate system federation

• Coordinate system federation
• Compute transforms between adjacent systems
• One system elects itself master and accretes
  adjacent systems by computing transforms into its
  space.
• Systems adjacent to the accreting system join by
  composing a transform into the accreting system.
• The process continues to grow the system until
  the error from successive transforms becomes too
  great

17
Computing coordinate transforms

• Each sensor that has coordinates in both systems
  can compute a coordinate transform
• These transforms are averaged
• A test point in ABC is transformed by each
  candidate transform
• The resulting cluster in DEF is averaged and
  back-substituted

B
A
D
E
C
F
18
Correlated error from long paths

• One problem for acoustic ranging is long paths
  due to obstructions
• The red obstruction affects the black sensors,
  which both measure a longer path to A
• The resulting cluster is bi-modal

B
A
D
E
C
F
19
Possible solutions to long paths

• Detection or correction of bad transforms based
  on
• analysis of clustering
• consistency checks across different compositions
  of transforms
• Comparison with other sensor modalities
• Selection of coordinate systems to eliminate bad
  transforms