building extraction and population mapping using high resolution images

Post on 14-Jan-2016

51 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

DESCRIPTION

BUILDING EXTRACTION AND POPULATION MAPPING USING HIGH RESOLUTION IMAGES. Serkan Ural, Ejaz Hussain, Jie Shan, Associate Professor Presented at the Indiana GIS Conference 2010 {sural,ehussain,jshan}@purdue.edu Tel: 764-494-2168 School of Civil Engineering Purdue University Feb 24, 2010. - PowerPoint PPT Presentation

TRANSCRIPT

BUILDING EXTRACTION AND POPULATION MAPPING

USING HIGH RESOLUTION IMAGES Serkan Ural, Ejaz Hussain,

Jie Shan, Associate Professor

Presented at the Indiana GIS Conference 2010

{sural,ehussain,jshan}@purdue.eduTel: 764-494-2168

School of Civil EngineeringPurdue University

Feb 24, 2010

AcknowledgementAcknowledgement

– Images and elevation data: Indiana View

– Building footprints, address data, and zoning

maps: Tippecanoe County GIS

– Census population data: U.S. Census Bureau

2

OutlineOutline

• Objective

• Population Mapping

• Study Area and Data

• Methods

• Assessment

• Conclusion

3

ObjectiveObjective

4

• Urban land cover mapping, especially buildings from high

resolution imagery and additional geospatial data using object-

based image classification

• Investigate the applicability of extracted building footprints as a

basis for micro-population estimation by disaggregation of

population at individual building level

Population MappingPopulation Mapping

5

• Estimation of population distribution at high spatial

and temporal resolution is of importance for

applications which use spatio-temporal distribution of

population together with other physical, social and

economic variables

• Public health

• Environmental health

• Urban planning

• Crime mapping

• Emergency response planning etc.

Population MappingPopulation Mapping

6

• Census

- once in every 10 years

- population reported of aggregate zones (e.g.

census blocks)

- predictions reported annually in township level

• Estimation of population at finer scales

- single housing and apartment units

• Mapping residential buildings from high resolution

images

Study Area and DataStudy Area and Data

7

• West Lafayette, IN

• CIR aerial images (2005)

- Resolution-1 m, 4 bands

• Elevation Data (digital elevation and surface

models)

- Resolution-5 feet

• Building footprints (2000)

• Building address points data

• City zoning map (scanned)

• U.S. Census 2000 population (census block level)

8

Test DataTest Data

CIR 2005 Aerial Image DSM

9

Test DataTest Data

Zoning Map-Scanned Zoning Map-Digitized

Residential planned development

Single family residential Single, two and multi-family residential

Non-residential planned development

Neighborhood business

10

Test DataTest Data

Address Point DataBuilding Footprints

Building ExtractionBuilding Extraction

• Availability of high resolution images (1 m)

– More details of ground objects

• Urban feature complexity

– Different objects with spectral similarity ( Roads, parking lots, walkways,

and building roofs)

– Similar objects with variable spectral response (Multi color roofs,

concrete and bituminous based impervious surfaces)

– Similar objects with a variety of shapes and sizes (buildings)

– Tree or their shadows covering houses, roads and street

11

High Resolution Images and Urban Features Complexity

12

Building ExtractionBuilding Extraction

• Object based image classification

- Segmentation: Division of image into homogeneous regions

– Classification:

Nearest Neighbor

Fuzzy rules (membership functions)

– Use of spectral, contextual and texture features for

classification

– Sequential classification

13

• Building extraction within census block group boundaries

14CIR 2005CIR 2005

Building ExtractionBuilding Extraction

Building ExtractionBuilding Extraction

• Land cover classification

15

Water Buildings (1, 2)

VegetationRoads Parking Lots

Shadow

TreesGrassResidentialNon Residential

Class hierarchy

Multi-family house

Single family house

General Business

Apartments

Classification ResultsClassification Results

16

Classification ResultsClassification Results

17

Classification ResultsClassification Results

18

• Height information (nDSM) derived from Elevation data

(DSM – DEM) for separation of elevated and non elevated

objects

• Zoning maps for the categorization of residential and non

residential buildings

• Use of address point data to check and validate the

classification of multi family houses based on building

(footprints) covered area

19

Classification Results – Classification Results – BuildingsBuildings

Classification Results – Classification Results – BuildingsBuildings

20

• Multifamily houses with less cover area mix up with

some of the single family houses with large

footprints

• Address point data can help to separate and

correctly classify residential buildings as single and

multi family houses

21

Classification Results – Classification Results – BuildingsBuildings

22Single family Houses Correctly classified -Multi family houses Misclassified -Multi family houses

Classification Results – Classification Results – BuildingsBuildings

• Buildings change detection between year 2000 and

2005

• Comparison of county building 2000 footprints with

buildings extracted from 2005 high resolution images

23

Classification Results – Classification Results – BuildingsBuildings

24

NO CHANGEDEMOLISHED

MISSED

NEW BUILT

2000 Building Footprints (County GIS) 2005 Building

Footprints (Image Classification)

Classification Results – Classification Results – BuildingsBuildings

25

Tract Type Missed False New Demolished

51 Business 1 - 3 5

52Residential 12 2 23 30

Business 1 - 6 3

Classification Results – Classification Results – BuildingsBuildings

Classification Results - BuildingsClassification Results - Buildings

26

• Buildings extracted from frequently acquired high

resolution images using object based classification

techniques may be suitable to be used as

supplementary data for

• Urban planning and development

• Monitoring urban growth/sprawl

• Maintaining and updating GIS building layers used

for various purposes etc.

Identification of Residential Identification of Residential BuildingsBuildings

27

• Disaggregate population at individual building level

• Distribute census population to the residential

buildings

• Filter out the non-residential buildings from

initially classified extracted building footprints

• Use different weights for different building types

• Refine the classification of buildings as houses and

apartment buildings

28

Building extraction

Small area non-residential building filtering using address points

Area threshold determination for small area non-residential buildings

CIR images

Filtered small area non-residential buildings

Address points

Small area non-residential building filtering using area threshold

Building footprints

Remaining building footprints

Remaining building footprints

Zoning maps

Residential / non-residential building classification

Non-residential buildings Residential buildings

Classify single family and apartment buildings

Google Maps & Site Visits

Address Points

Identification of Residential Identification of Residential BuildingsBuildings

Identification of Residential Identification of Residential BuildingsBuildings

29

Residential planned development

Single family residential Single, two and multi-family

residentialNon-residential planned developmentNeighbourhood business

Zoning MapZoning Map

Identification of Residential Identification of Residential BuildingsBuildings

30

Address DataAddress Data

Dasymetric Mapping of Dasymetric Mapping of PopulationPopulation

31

Areametric: Volumetric :

Weighted Areametric:

Weighted Volumetric:

Building population

Census unit population

Building Area(Lwin and Murayama, 2009)

Weighting factor

Building Volume

2000 Census Population 2000 Census Population Distribution Distribution

32

2000 Census Population 2000 Census Population Distribution Distribution

33

RMSE (2000)Method n = 89 n = 84

Areametric

56.81 51.05

Weighted Areametri

c30.20 23.31

Volumetric

43.79 38.69

Weighted Volumetri

c34.38 24.34

U.S. Census Population U.S. Census Population PredictionsPredictions

34

• Building footprints extracted from 2005 high

resolution images

• U.S. Census Bureau provides annual predictions at

township level

• Extent of the study area is a subset of a township

• Trend of population change modeled by fitting a 5th

order polynomial to U.S. Census predictions at

township level

• Obtained trend is used to obtain the population of the

census blocks in the study area at 2005

U.S. Census Population U.S. Census Population PredictionsPredictions

35

Year

US Census

Predicted Populatio

n

Population Growth Rate (%)

2008 54691 1.1582007 54065 1.3632006 53338 2.2642005 52157 0.8922004 51696 -0.0102003 51701 -0.0972002 51751 1.0172001 51230 -0.0062000 51233

2005 Predicted Population 2005 Predicted Population Distribution Distribution

36

AssessmentAssessment

37

• Tree cover

• DSM errors

• Census data problems

- Census block boundary alignment

- Non-correspondence with existing residential

buildings

• Data integration

DSM ErrorsDSM Errors

38

Census 2000 Data ProblemsCensus 2000 Data Problems

39

Census Block #4001

Census Block #4000

Number of Residential Buildings

= 20

Census 2000Population

= 3Census 2000Population

= 51

Number of Residential Buildings

= 1

ConclusionsConclusions

40

• Object based image classification is an effective

method to extract buildings from high resolution

images

• Integration of elevation data further improves

building extraction

• 98% overall classification accuracy achieved using

both high resolution images and elevation data

• Volumetric method produce better results than

areametric method without the inclusion of a

weighting factor

• Inclusion of a weighting factor improves the results

for building population estimation

• Further classification of building types may improve

the estimation results

41

top related