LOS TWG background

1
Line-of-Sight (LOS) Analysis and Algorithms A
Historical Perspective
(Also referred to as Line-of-Sight/
Intervisibility/ Viewshed) 5 June 2003
Danny C. Champion [phone]
champd_at_trac.wsmr.army.mil
2
Disclaimer
The information in this briefing is probably not
100 percent correct. It is correct from my frame
of reference. One of the purposes of todays
discussion is to create a more accurate
representation of the history of LOS and the
digital terrain data used by the algorithms.
3
Historical Factoid

• The M1/M2 were developed using a terrain box
  containing 60x66 data points at 100m resolution
  (6x6.6km)
• The next generation of combat vehicles will use
  a terrain box of 2000x3000 data points at 25m
  resolution (50x75km)

Data Volume is increasing
Computer Power is increasing
4
Agenda

• Things to be covered
• Definitions
• Purpose of LOS Algorithms
• How Terrain is Depicted
• How the Algorithms Work
• Algorithm Considerations
• Users Considerations
• Some Questions for NIMA
• Ideas for Increasing Algorithm Speed
• General Education Concepts for Users
• LOS VVA
• History
• Emerging/Pending Work
• VVA Metrics
• Things not covered
• Nuts and Bolts (Save for Another Briefing)

5
Definitions
Line of Sight is a function of geometry. Detectio
n is a function of sensor type, contrast, visible
light, sky over ground ratio, range, atmospheric
attenuation, the exposed area of target, and
whether or not the target is moving.
6
What Do You See? Is this Just A Tree?
7
Perhaps!
8
Original Algorithms

• About 1960, several places independently
  developed LOS algorithms
• BRL, BDM, SRI, APG, IITRI, Ohio State,
• USNA
• Data was manually extracted from military maps
  (about 8 mapsheets per year)
• Data was first plotted using Calcomp software

9
Purpose of LOS Algorithms

• Simulated LOS for Models and Simulations
• Position Sensors to Maximum Visibility
• Position Targets to Minimize Visibility
• Position Industry to Minimize Visibility
• Insure Microwave Towers are Visible to
• Each Other
• Basically, we need to answer the question
• Can a sensor at location A see a Target at
• location B?

10
How is the Terrain Depicted? Everything
Starts as a Grid Grid Grid Cells TINs Splines Oth
er Methods (i.e. contour lines)
Each Representation Can have a Family of
Algorithms!!
11
Grid Representation of Terrain

• Each post represents
• an elevation.
• Algorithms use different methodologies to fill
  in terrain between posts.
• Math is easy.

• Janus
• DYNTACS
• ModSAF
• ALBE
• SEEFAR

Most Common Representation
Most Accurate
12
Cell Terrain Representation

• Each grid is assumed to be a plateau.
• Math is easy.
• Usually less accurate.

• Bresenham
• Binary Search Algorithm

Very Fast !!
13
Terrain Represented as a Triangular Irregular
Network

• Complex Math
• Reduced Storage
• Variable Resolution
• Many Methodologies
• to TIN Terrain (dozens)
• with varying accuracy
• Visual Simulators
• Not Differential at Edges

As TINs are reduced, what happens to LOS?

• Exact Geometric Intersection
• Sample Based Estimation

14
Terrain Represented as a Spline

• Immature
• LOS computed as
• a directional
• derivative
• Very Slow

15
Other Terrain Representations

• Fourier Series/Chebychev Polynomials
• Wavelets
• Fractals
• Mixtures
• DYNTACS/SEEFAR
• Splined TINs (to smooth out edges)

Untested
16
Ground Rules for Discussing LOS Algorithms
We will examine concepts, not
proprietary software implementation
Example LOS in a TINned Database Using the
Exact Geometric Intersection Method
Concept LOS is examined at all of the
intersections between the polygon edges and the
sensor/target line
(Continued)
17
Ground Rules (Continued)
Implementation 1 Put polygons into a list and
examine every edge of every polygon in the
database when calculating LOS. Implementation
2 Degenerate edges of polygons into an edge file
and examine intersection of edges between sensor
and target for blockage
Same Answer Different Implementations!!
18
Algorithm Considerations

• Coordinate System (UTM or Lat/Long or
  Geocentric)?
• Curved or Flat Earth or Radar LOS?
• How is sensor/target elevations determined?
• How are features (vegetation/urban) treated?
• How is LOS blocked? (slope or calculated
  elevation)
• Terrain Resolution and its effects on LOS
• Lower Resolution faster, less accurate
• Higher Resolution slower, more accurate
• How should the algorithm be implemented?
• Which algorithms work best with my hardware?
• Precision, Precision, Precision

19
Algorithm Considerations (Continued)
16m
14m
What is the elevation of ?

• 13.52 Bilinear Interpolation
• .00 Nearest Point
• 10.00 Default to SW Corner
• 14.80 Left Hand Triangle
• 13.20 Right Hand Triangle
• 13.35 Weight Points

10m
12m
Which method should I use? Which is right?
20
Algorithm Consideration (Continued)
How is LOS in vegetation calculated?
LOS Blocked by Vegetation
LOS Partially Blocked by Vegetation
21
Precision Example

• DYNTACS LOS Algorithm
• 35646 Challenging, Tactical LOS, 5m Resolution
• Three Implementation - Mathematically Equal
• Original DYNTACS (steps across lattice)
• Point/Slope Intersection Method
• Point/Slope Intersection Method (Double
  Precision)
• DYNTACS versus
• Point/Slope 36 differences
• Point/Slope (double precision) 4 differences
• Two Methods for Blocking LOS No Difference
• Slope
• Elevation

22
User Considerations

• How will each of these depictions effect
  visualization?
• How do I define best when calculating LOS?
• Fast (but less accurate)
• Accurate (but slower)
• Reciprocal (if I can see you, can you see me?)
• How do I match algorithm and the accuracy of
  available data?
• What is the best algorithm for my hardware?
• What resolution of terrain should I consider?
  Again!
• Do I know the difference between LOS and
  Detection?

23
Some Questions for NIMA

• What is the source and extraction method of the
  elevation data?
• How is data accuracy going to change over the
  next few decades?
• How should I use SRTM reflective surface data?
• How should I handle the gaps in SRTM data?
• How will other source data (different
  IFSAR/LIDAR/) affect DEM accuracy?

24
Ideas for Faster LOS

• LOS on a chip or parallel processing
• Finding faster methods for finding the
  interruption of LOS
• Approximations (more loss of accuracy)

25
Educating the User/Programmer

• Datum
• Ellipsoids
• Geoids
• Data Sources
• NIMA
• USGS
• Many Other Sources

26
LOS Algorithm VVA

• Completed Studies
• Emerging/ Pending Work
• VVA Metrics

27
Completed Studies
TETAM 1974 (Combined Arms Center)  Field
Experiments compared LOS to DEMs Dragons Eye,
Kings Ride, Chinese Eye early 1970s
(UK/TRAC)  A-Station Location Analysis for Fort
Irwin and Fort Sill 1980 and 1981 (TRAC)  TEC
Study on Terrain Requirements (Tactical Terrain
Data) 1984?  Tactical Terrain Intervisibility
Classification Study October 1986
(TRAC)  Digital Terrain Elevation Data
Resolution and Requirements Study - November 1990
(TEC/TRAC)  The Effects of Digital Terrain
Elevation Data Representation on Combat
Simulations December 1993 (TRAC/TEC)  The
Effects of Different Line-of-Sight Algorithms and
Terrain Elevation Representations on Combat
Simulations September 1995 (TRAC/TEC)  Yuma
Terrain Evaluation August 1997
(TRAC/TEC)  Fort Benning High Resolution Terrain
Evaluation May 1997 (TRAC/TEC/IDA)  LOS in
Vegetation Studies 1998-2001 (TRAC/TEC)  Line-o
f-Sight Field Validation Exercise Sponsored by
Operational Test Command (OTC), Fort Hood June
2002 (TRAC/TEC)
Only a Partial List
28
Emerging/Pending Work

• Evaluating SRTM data
• Goose Bay Evaluation (SRTM, other IFSAR)
• Monument Valley Evaluation
• Fort Bliss Evaluation

29
Current Metrics Used for VVA

• LOS Agreement
• LOS Agreement Sensitivity
• Analysis of Disagreements
• Terrain Profiles
• Evaluating Between Real Terrain and Algorithm or
  Between Two Algorithms
• Sensitivity of Sensor/Target Heights

30
A Few VVA Results
31
Original DEM (4m resolution and DYNTACS
algorithm)
32
Two Meter Rise Missed By First DEM
33
LOS Analysis of TINs (Compared to Original 4m
TRN 2,000,000 TINs)
Number LOS of
TINs Agreement 1,996,238
0.992 221,447
0.988 55,186
0.979 30,974
0.979
34
4m DEM and 32m TIN
One Time Example
1,002,001 data points Versus 15,625 data points
35
Percent Agreement of Other LOS Algorithms
Compared to DYNTACS
Grided TINned
Spline Algorithms
Algorithms Algorithms
L1-Splines 14641 Posts 98.57 1681 Posts
95.65 441 Posts 90.47 Conventional 14641
Posts 98.72 1681 Posts 97.59 441 Posts
88.59
7200 Posts 98.35 3600 Posts 97.00 2400 Posts
96.55 1440 Posts 96.55 960 Posts 93.69 720
Posts 94.89
Janus (Old) 99.25 Janus (New) 94.74 Bresenham
93.47 ModSAF 96.77
1332 Points 29 Palms, CA
36
". . . Use intervisibility statistics with great
care. No matter if they are taken from field
measurements or computer simulation or other
sources, the basis for the findings must be well
understood, or the user's analysis is at
peril." - Warren Olson MORS Handbook on
Terrain, Unit Movement, and Environment, April
1994