DLA CONFERENCE 20136‐8 June | Bernburg, Germany
June 7, 2013
Photo‐based Terrain Data Acquisition & 3D Modeling
Howard Hahn Kansas State University
Partial funding by:KSU Office of Research and Sponsored Programs
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Need
Application 1 – Monitoring Gully Erosion
Source: H. HahnMcPherson County
• Field Erosion Progression
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Need
Application 1 – Monitoring Gully Erosion • Manual transects performed using GPS (x,y) and laser level (z)
Source: H. HahnMcPherson County
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Need
Application 1 – Monitoring Gully Erosion • Manual transects (~15) require one day each major rainfall
Source: H. HahnMcPherson County
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Need
Application 1 – Monitoring Gully Erosion Terrain Modeling Needs
• Reduce time to perform transect measurements
• Preserve a photographic and 3D model record for periodic field monitoring
• Estimate the volume of erosion taking place
• Integrate the 3D model in a GIS environment to correlate with soil, slope, and drainage patterns
Source: H. HahnMcPherson County
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Need
Application 2 – Micro Drainage Modeling
Source: D. CrossKansas State University
• Residential neighborhood being threatened by eroding creek
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Need
Application 2 – Micro Drainage Modeling
Source: D. CrossKansas State University
• Analyzing drainage patterns between structures
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Need
Application 2 – Micro Drainage Modeling• Proposed water treatment chains and detention areas
Source: D. CrossKansas State University
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Need
Application 2 – Micro Drainage Modeling
• Terrain map post construction sites for design of stormwater best management practices (BMPs)
• Schedule flexibility to fly anytime, anywhere (not currently possible without COA)
• Aerial photo acquisition for small area sites where it is not cost effective to fly LiDAR
• Map drainage networks at potentially higher resolution than what is available through county 2m LiDAR data
Terrain Modeling Needs
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Tools (High End)
Current Automated Techniques
http://www.wy.nrcs.usda.gov/wygis/lidar.html
• Aerial LiDAR platforms
Typical Specifications
‐ Lateral placement accuracy: 10‐30 cm‐ Vertical placement accuracy: 7‐16 cm‐ Filter first returns for bare earth results‐ Platforms: Plane or helicopter
Examples: Leica ALS70‐cm
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Tools (High End)
Current Automated Techniques• Digital Photogrammetry
Typical Specifications
‐ Optical/sensor: DTM and ortho from same source‐ RGB color, color infrared, & panchromatic‐ On‐the‐fly L1 image rectification‐ Triangulation for DTM: Vert +/‐ 8 cm; contour 15 cm‐ Ground sample distance (GSD): 5‐25 cm
Examples: Leica ADS80, Aerometrex UltraCam
http://commons. wikimedia.org/wiki/File:Geo‐Referenced_Point_Cloud.JPG
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Tools (High End)
Current Automated Techniques• Terrestrial LiDAR
Commons.wikimedia.org/wiki/file:Lidar_P1270901
Typical Specifications
‐ Range: 120 m‐ Target acquisition: up to 50 m‐ Point resolution: 3 mm at 50 m‐ Speed: 1 million pps
Examples: Leica ScanStation P20
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Introduction: Simpler options
• Relatively low cost• Rapid and easy field
operation• Sub‐dm accuracy
Goals
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Software Selection
Agisoft PhotoScan Pro ‐ Principle• Photogrammetric bundle triangulation
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Software Selection
Agisoft PhotoScan Pro Applications• Generation of digital elevation models & orthophotos
3D Point Cloud 3D Triangulation Ortho Mosaic
Source data by Dr. K. PriceAgisoft Model by H. Hahn
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Software Selection
Agisoft PhotoScan Pro Applications• Area & volume measurements
Stockpile Displacements Mining Excavations
http://downloads.agisoft.ru/photoscan/sample04/sample04.pdf
http://downloads.agisoft.ru/photoscan/sample04/sample05.pdf
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Software Selection
Agisoft PhotoScan Pro Applications• 3D model reconstruction (archaeology, etc.)
Agisoft model by H. Hahn
Adapted from http://www.voicesfromthedawn.com
/kilclooney‐dolmen/
Kilclooney Dolmen (Scotland) 3D texture mapped model
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Photo Collection
Terrestrial • Standard 35 mm DSLR (Nikon D50)
• Photographic parametersCamera calibration derived from extracted digital photo info (FOV, FL, etc.) Recommended 5 megapixel or higherMove camera position60% side overlap; 80% lateral overlap50‐80%At least two scale markersMaintain 45‐90°angle to subject
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Photography Collection
Airborne Systems • In the USA, subject to regulation by the Federal Aviation
Administration and requires Certificate of Operation (COA)• Unmanned Aerial Systems (UAS) standards by 2015• Camera: Lightweight Canon Powershot S100 (wide angle)
UAS vehicles and photos by Dr. Kevin Price, KSU
RiteWing RC Zephyr Flying Wing
DJI S800 Hexacopter
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Methods
Test Site #1 – Open Field• Purpose: To determine if drainage patterns could be
modeled in sufficient detail to replace LiDAR
COA certificatePlatform: Zephyr wingImages used: 9Area: ~ 12.9 haCover: short grassGround control: 2m LiDARand Bing Imagery – 16 pts
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Methods
Flight path and composite photo match
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Results
Test Site #1 – Open Field• Traditional LiDAR yields superior results
Open Field Open Field
Road trace
Defined drainages
Leafless tree patch (typ)
Evergreen tree patch (typ)
Control point.(typ)
• Inadequate photos per flight speed• No bare earth processing (affected by vegetation)
2m LiDAR Agisoft PhotoScan 3D Model
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Methods
Test Site #2 – Prairie Gully• Purpose: To test if very low altitude aerial imagery/
photomodeling can replace manual transect techniques
COA certificatePlatform: DJI HexacopterImages used: 16Area: ~ 800 m2
Cover: grass/dirtGround control: 2m LiDARand Bing Imagery – 3 pts
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Methods
DJI Hexacopter Flight Path
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Methods
3D Point Cloud
Triangulated surface model
Texture mapped 3D model
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Results
Test Site #2 – Prairie Gully• Overall accuracy within 3 cm
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Methods
Test Site #3 – Eroded Slope• Purpose: To test if Agisoft could resolve minute surface
differences enabling erosion volume calculations
Handheld cameraImages used: 9Area: ~ 6.7 m2
Cover: dirtGround control: 2 markers
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Methods
Triangulated surface model
3D Point Cloud
Texture mapped 3D model
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Results
Photo Series & Surface #1 Photo Series & Surface #2
Test Site #3 – Eroded Slope• Error Control – Photographed surface x 2
Change detection noise
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Results
Photo Series & Surface #1 Scraped Surface
Test Site #3 – Eroded Slope• Error Control – Photographed surface x 2
Change detection noise
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Conclusions• TS 1‐ Field
• Matched GIS control points within +/‐ 1.5 m
• Terrain produced via Zephyr flying wing and PhotoScan did not yield better results than 2m LiDAR
• Too few photos acquired due to long photo interval relative flight speed (~ 60 km/hr) contributed to lower quality
• Vegetation, which could not be filtered from point cloud, contributed to lower quality compared to LiDAR results
• TS 2 – Prairie Gully• Imagery acquired from hexacopter flying at low 16m altitude
used to generate 3D terrain model with a mean error of 3.53 cm compared against manual measurements
DLA CONFERENCE 20136‐8 June | Bernburg, Germany
Conclusions• TS 3 ‐ Eroded Slope
• At close photo ranges, PhotoScan is capable of resolving minute terrain changes (+/‐ 5 mm)
• For a small area test, Initial investigations demonstrated that accuracy of at least 85% can be achieved when estimating erosion quantities.
• Future Research• Re‐test drainage network mapping for improved accuracy. For
smaller fields, switch to a hexacopter aerial platform at slower flight speed. Attempt to filter vegetation from point clouds
• Create field gully models over an extended monitoring period, align models, and estimate erosion quantities