The Devil is in the Details:

1
The Devil is in the Details Simulation Databases
for Video Surveillance
David McKeown, Mike Polis, Russ Schaaf, Steven
Sprang TerraSim, Inc. Pittsburgh, PA Joseph H.
Findley III Synthetic Environments Program TRW,
Inc. Alexandria, VA Nancy Gardner U.S. Army
Topographic Engineering Center Alexandria, VA
2
Video Surveillance and Monitoring(VSAM)

• Defense Advanced Research Projects Agency (DARPA)
  sponsored project within Image Understanding
  Program
• Develop a cooperative, multisensor, video
  surveillance system to provide continuous
  coverage over urban and battlefield environments
• Integrate isolated video stream for understanding
• humans (individual, groups), vehicles
• With real world geospatial activity models
• activity prediction, activity hand-off

3
VSAM Database Requirements

• Two test sites Schenley Park (CMU Campus)
• Fort Benning (MOUT
  Site)
• Requirements for high geospatial accuracy (sub
  meter)
• Must correlate with the live exercise for virtual
  playback
• Provide a synoptic view of complete set of
  activities
• Modeling of Video Sensors to provide
• 3D position
• sensor instantaneous field of view
• sensor pan/tilt field of view
• predictable line of sight
• Conversion to MODSAF CTDB

4
VSAM Virtual World Goals

• Provide highly accurate virtual sight models to
  support video understanding researchers
• Support the development of virtual camera models
  within MODSAF
• Virtual camera allows for video algorithm
  development, testing, planning for video sensor
  deployment
• Replay of video streams vs. virtual world
  movements.

5
VSAM Database Construction

• Construction using non-traditional geospatial
  data sources
• Incorporate a high degree of automation
• Avoid manual modeling using TerraTools
• Integrated TIN (ITIN) representation
• Integrated features (roads, sidewalks, drainage,
  bridges)
• Heterogeneous data sources
• GIS data, ground survey, NIMA source, aerial
  imagery
• Building model sources
• As-built building models (Openflight, AutoCad)
• Generate representative buildings from GIS
  footprints
• Photogrammetrically derived from aerial imagery

6
The Schenley Park Database
7
ARC/INFO GIS Source Data
Detail of GIS Data with 5-foot contours
GIS Source Data
8
Schenley Park Statistics

• 415 Acre park with internal trails, paths, and
  roads
• Dense transportation network, bridges and unique
  building architectures
• 3x3 km total area
• 115.8 km of roads
• 5995 unique building structures (4265 residential)

9
Schenley Park Buildings

• Building Area/ Footprint
• Derived from GIS source data
• Building Heights
• Proportionally derived from building area
• Building Roofs
• Peaked or flat roofs were assigned based on
  other building attributes such as size and shape
• Residential buildings were assigned peaked
  roofs commercial buildings were assigned flat

10
CMU Campus
11
GPS Survey Points and Camera Locations
12
Video Tracking
Live view for video tracking
Video View tracking
13
Accuracy of Schenley Park Database

• Assessed by measuring building sides and corner
  to corner distances
• On 15 building segments
• RMS Error 1.4 m
• Average error relative to length 5
• Of 8 intrabuilding distances (corner to corner)
• RMS Error 1.8 m
• Relative error 4.5

14
Accuracy of Schenley Park Database

• Assessment of vertical accuracy of video camera
  points
• Point A Camera Position at eastern peak of PRB
  Building

15
Accuracy of Schenley Park Database

• Assessment of vertical accuracy of video camera
  points
• Point C Wean Hall camera position along north
  face of building

16
The Fort Benning Database
17
Source Data Ft. Benning
15000 Scale Imagery
18
Source Data Ft. Benning
Contours and Trees
Digitized Areal Features
19
Source Data Ft. Benning
OpenFlight Building Models
Road Network (DXF format)
20
The Ft. Benning DatabaseStatistics

• 17 previously constructed building models
• 17.7 miles of type-specific roads
• 260 Geospecific tree models
• 45 additional models derived from an orthophoto
• fire hydrants
• manholes
• railroad ties
• telephone poles

21
The Ft. Benning DatabaseTree-Model Placement

• Eastern side of MOUT
• GPS tree positions compiled by land survey team
• TerraTools automatically placed models into
  terrain
• Western side of MOUT
• Tree positions determined by orthophotos
• Models hand edited into terrain using TerraTools
  tedit

22
McKenna MOUTMilitary Operations on Urban Terrain
23
Ft. Benning Tree Models
24
Ft. Benning Tree Models
25
Imagery vs. Simulation
26
Imagery vs. Simulation
27
Imagery vs. Simulation
28
Imagery vs. Simulation
Limitations of Simulation Database Construction
29
Accuracy of Ft. Benning Database

• Models were placed using intrabuilding distances
  (corner to corner)
• Residual Error Analysis

30
VSAM Database Complexity
31
Area Normalized Polygon Count Comparison of
Databases
32
Area-Normalized Polygon Count Comparison of
Databases
33
Polygon Count Comparison of Databases
34
Conclusions

• DARPA VSAM Program requirements for urban
  databases
• Unique in terms of requirements for
• Geodetic accuracy and completeness
• Use of hetergeneous (and non traditional)
  datasources
• As-build measurement and validation of database
  accuracy
• Support of real time video stream analysis in
  MODSAF
• Successful
• Use of COTS automated database construction tool
• TerraTools V1.1
• Several iterations of database construction with
  USATEC
• Timely delivery of three VSAM databases