smartboxes for interactive urban reconstruction
DESCRIPTION
SmartBoxes for Interactive Urban Reconstruction. Liangliang Nan 1 , Andrei Sharf 1 , Hao Zhang 2 , Daniel Cohen-Or 3 , Baoquan Chen 1 1 Shenzhen Institutes of Advanced Technology (SIAT), China 2 Simon Fraser University, Canada 3 University of Tel Aviv, Israel. 3D Cities. - PowerPoint PPT PresentationTRANSCRIPT
SmartBoxes for Interactive Urban Reconstruction
Liangliang Nan1, Andrei Sharf1, Hao Zhang2, Daniel Cohen-Or3, Baoquan Chen1
1 Shenzhen Institutes of Advanced Technology (SIAT), China2 Simon Fraser University, Canada
3University of Tel Aviv, Israel
Virtual BerlinVirtual Philadelphia
3D Cities
Acquisition of Urban Environments
Cameras/videos
Remote Sensing Systems
3D Digital CityAuto-mounted LIDAR
Airborne LIDAR
3D LiDAR scanner
• Street-level• 60km/h• 180 pitch• [100-300m] range • 100K points/second• 5cm XY accuracy
Outdoor Urban Scanning
Imperfect Scans - Occlusions
• Point cloud contains holes due to various occlusions (“shadows”)
Imperfect Scans – Angle & Range
• Oblique scanning angle• Laser energy attenuation
on range
Urban Building Characteristics
• Repetitions, intra symmetry and regularity• Axis-aligned basic primitives• Dominant planes
SmartBoxes
• Box-up and Smart!
SmartBoxes
• Box prior shape fitting • Smart context awareness
Both
Context
Data
Live Demo
9
Related Work
• Procedural modeling of buildings and facades [Wonka2003;Muller2007]
• Automatic 3D reconstruction from 2D images [Zisserman2002;Xiao2009;Furukawa2009]
• Interactive modeling of architectural structures [Debevec1996;Schindler2003; Xiao2008;Sinha2008;Jiang2009]
• Primitive fitting to data [Gal2007;Schnabel 2009]
Preprocessing
• Automatic detection of planes and edges assuming dominant orthogonal axes– RANSAC planes– Line sweep edges
Snapping a Box
• 2D rubber band ROI• Collect planes, edges, corners• Find the best fitting box using
data fitting force D(B,P)
Data fitting D(B,P) snap
Facet Edge
• Data fitting force
Data quality(Confidence + Density)
Distance
Grouping
• Simple SmartBox Compound SmartBox• Align to remove gaps and intersections
– cluster and align close to co-linear edges
Drag-and-drop context C(Bi-1, Bi)
• The context of Bi
••• Bi-1Bi-2Bi-3
Interval Alignment Scale
Context
Bi
Drag-and-drop context C(Bi-1, Bi)
• The context of Bi
– Interval term
••• Bi-1Bi-2Bi-3Bi
Drag-and-drop context C(Bi-1, Bi)
• The context of Bi
– Alignment term
••• BiBi-1Bi-2Bi-3
Bi
Drag-and-drop context C(Bi-1, Bi)
• The context of Bi
– Scale term
••• BiBi-1Bi-2Bi-3
Discrete objective minimization
• Find linear transformation T(excluding rotation) to minimize:
Data fitting force
Contextual force
Discrete objective minimization
••• Bi-1Bi-2Bi-3
• Find linear transformation T(excluding rotation) to minimize:
Data fitting force
Contextual force
Balance between two forces
Emerging city of Shenzhen, China
Results: textured buildings
Manchester Civil Justice Centre (Manchester, UK)Habitat 67 (Montreal, Canada)The Crooked House (Sopot, Poland)
(Thank You)!