Val Noronha University of California, Santa Barbara Centerline Extraction and Road Condition.

Val NoronhaUniversity of California, Santa Barbara

Centerline Extraction and

Road Condition

Asset Management 2001-09-23 #2 N C R S T

Why centerlines?

Accurate (x,y) for• ITS• precision applications• location based services

Accurate length for compatibility with linear referencing

Asset Management 2001-09-23 #3 N C R S T

Centerline Applications

Application Scale

Market research 1:50KEmergency reporting 1:20KHighway operations 1:20KLane tolls 1:5KSnow plowing 1:1K

Asset Management 2001-09-23 #4 N C R S T

Approaches to deriving centerlines

Convert old maps

Convert new maps, integrate CAD plans

Photogrammetry GPS

Asset Management 2001-09-23 #5 N C R S T

Outline

Centerlines from GPS Centerlines from hyperspectral

imagery Other uses of hyperspectral

analysis: early findings on road condition

Asset Management 2001-09-23 #6 N C R S T

GPS for Hwy Ops

Hi end

Lo end

Asset Management 2001-09-23 #7 N C R S T

Low-end GPS units

$250 $195 $150

Asset Management 2001-09-23 #8 N C R S T

Lane Discrimination Test

Asset Management 2001-09-23 #9 N C R S T

Lane Discrimination Test

Asset Management 2001-09-23 #10 N C R S T

Lane Discrimination Test

Asset Management 2001-09-23 #11 N C R S T

The one to beat …

$150 at CompUSA

Convenience Price Can RS beat

this?

Asset Management 2001-09-23 #12 N C R S T

Remote sensing centerline strategy

Find pixels that represent road … hyperspectral library

Detect linear patterns, form centerlines

Attach legacy attributes

Compare costs and benefits

Asset Management 2001-09-23 #13 N C R S T

3-step hyperspectral process

MESMA Q-tree Vectorize

Additional steps: clean, revisit, conflate

Easy Street New neighborhood Little or no foliage overhang Vehicles in garage/driveway

Not so easy Repairs and surface coats Paint stripes Shadows Parked vehicles Foliage overhangs

Asset Management 2001-09-23 #16 N C R S T

Multispectral sensorsR

eflect

ance

400 700 20 20 50 40

Infra-red

Wavebands originally optimized to sense health of Soviet wheat

Asset Management 2001-09-23 #17 N C R S T

Hyperspectral sensorsR

eflect

ance

400 700

20 30 50

… 2400

Each pixel is characterized by 200+ reflectance values

20 30 20 30 5020 30 20 30 5020 30 205030

Asset Management 2001-09-23 #18 N C R S T

Hyperspectral road identification

Materials have unique hyperspectral signatures, based on chemistry, texture, etc

What are the principal materials found in roads … what are their signatures?

Study them at close range in the field (handheld spectrometer)

Then see if you can detect the signatures from imagery (4m airborne AVIRIS by JPL)

Asset Management 2001-09-23 #19 N C R S T

ASD full range spectrometer

Field Spectrometer

Concretes

0

0.1

0.2

0.3

0.4

0.5

350 850 1350 1850 2350

Asset Management 2001-09-23 #20 N C R S T

499 roof 179 road 66 sidewalk 56 parking lot 40 road paint 37 vegetation

Field Spectra Collected

47 non-photosynthetic vegetation (bark, dead wood)

27 tennis court 88 bare soil and

beach 50 miscellaneous

other urban spectra

Asset Management 2001-09-23 #21 N C R S T

0

0.1

0.2

0.3

0.4

0.5

350 850 1350 1850 2350

Old Concrete

New Concrete

Concrete Bridge

Red Tinted

Concretes

0

100

400 900 1400 1900 2400

0

100

200

300

400

400 900 1400 1900 2400

0

100

200

400 900 1400 1900 2400

Con

crete

roof

Park

ing

lot

Asp

halt

road

Asset Management 2001-09-23 #24 N C R S T

Step 1 result

MESMA

Asset Management 2001-09-23 #26 N C R S T

Step 2 result

Q-tree

Asset Management 2001-09-23 #27 N C R S T

Step 3 result

Vectorize

Asset Management 2001-09-23 #28 N C R S T

Step 3 result

Asset Management 2001-09-23 #29 N C R S T

Where it Fits in the Big Picture

Applications Scale Convert Remote GPS Photo- Enggpaper sensing grammetry plans

Mkt research 1:50K EMS 1:20K Hwy ops 1:20K Lane tolls 1:5K Plowing 1:1K ITS 2010 1:100

Global scale —Logistics

Local scale, esp urban —Asset mgmt

Asset Management 2001-09-23 #30 N C R S T

Road condition

Asset Management 2001-09-23 #31 N C R S T

Field Data Records

Asset Management 2001-09-23 #32 N C R S T

0

0.05

0.1

0.15

350 850 1350 1850 2350

Dry Oil

Wet Oil

Tar Patch

Sealcoat

Surface Treatments

Asset Management 2001-09-23 #33 N C R S T

0

0.05

0.1

0.15

0.2

0.25

0.3

350 850 1350 1850 2350

Old

New

Age

Asset Management 2001-09-23 #34 N C R S T

0

0.05

0.1

0.15

0.2

0.25

350 850 1350 1850 2350

Butte

Berkeley Good

Berkeley Bad

Good vs bad indistinguishablePixel size much larger than cracks and patches

Surface “Quality”

Asset Management 2001-09-23 #35 N C R S T

In conclusion …

RS for centerlines• a fully automated solution is not yet

here• potential for the future

RS for road condition• much promise

1

www.ncgia.ucsb.edu/ncrst