Automation & Visualization in Immersive Geographic Virtual EnvironmentsThomas J. Pingel, Keith Clarke, William McBride Department of GeographyUniversity of California, Santa Barbara
SCRAM Meeting, Wednesday, April 20, 2011
Central Research Question:How can an immersive geographic virtual environment assist in the interpretation, analysis, and understanding of specific, local events?
I will focus on issues of data, language, and design.
We are constructing an experimental testbed that merges lidar, video, and the usual geodata.
Up to now, weve mostly been working on the automation part. It is going well.
FilterTotal Error (%)Kappa(%)Axelsson(1999)4.884.2Silvan-Cardenas (2006)74.8Chen (2007)7.2Meng (2009)79.9SMRF4.186.4Current3.688.2
Why (or when) 3D?SuccessesGame Industry (US $20B per year) First Person ShootersCall of Duty: Black Ops -$360M on the first dayRole Playing GamesWorld of Warcraft - $1B per yearFlight and Battle SimulatorsGoogle EarthFailures3D avatar chatCNNs election-night tech showcasesStreet view (?)Compare Digitally Mediated Communication:Texting, Email, Telephone, Videoconference
Immersive Geographic Virtual EnvironmentsVirtual: Computer generated; 3D & TimeImmersive: Multiple Psychologies of Space (Montello, 1993)Figural VistaEnvironmentalGeographical We want to representing Environmental (or Geographical) spaces as Figural (or Vista) Objects while retaining some of the cognitive elements of each. (Which? How?)Emphasis on representing places in a model that can be used as an object or experienced as a place.
Google EarthWhy?Free client thats widely distributedReasonably intuitive user interfaceSupplied backdrops and easy overlaysWhy not?Interface is a virtual figural object representing a geographical spacePrefer figural / vista representation of an environmental spaceNo control over interfacePoor interaction with dataGreat for simple visualization, poor for analysis
Choosing A Platform for Geographic Virtual EnvironmentsGoogle Earth is ubiquitous, has plentiful ancillary data, and reasonably good visualizationGround navigation is poor, and it is difficult to modify the terrain and otherwise customize (i.e., the strength of GE can also be a weakness).Plenty of other virtual globe projects, but with mostly the same issuesESRIImprovements to ArcGIS 10 Scene and Globe packages. COLLADA is now supported.Easy to integrate data, more difficult to customize and create worldsCityGML is geography awareIntegration with other elements not so goodVTK (Visualization Toolkit)Efficient graphical languageLimitations in multimedia (sound, movie)Game engines offer excellent visuals & navigation, rich interaction, and full customizabilityUnity & Torque are popular choices, and the classic Unreal game engine supports COLLADAVery poor geographic supportX3D & COLLADA
Model LanguagesCOLLADAProposed by Sony, managed by Khronos GroupInterchange format.dae (digital asset exchange)Supported in Google Earth, SketchUpNominal geospatial componentX3DSuccessor of VRMLBacked by Web3D ConsortiumStrong support from Naval Postgraduate SchoolEmphasis on full scene creation & storageGeospatial component (successor to GeoVRML)X3D-EarthGoalsOpen sourceWide community supportGood documentation
X3D COLLADABoth are overlapping, open-source technologies to simplify construction of 3D objects and scenesIn practice, even translations between these formats is mediocreGaming engines favor COLLADA, as does Google Earth, but many other 3D browsers favor X3D and VRML.
X3D BrowsersXj3DJava, runs everywhereImplements most X3DNavigation and device support is poorInstantRealitys Instant PlayerBest Geospatial Component supportBest Navigation (Game / GeoExamine)Renders well above and on surfaceGeoOrigin ImplementationBitManagement ContactPoor on-ground visualization (jitter)Incomplete implementation of Geospatial Component Good device support, immersion, stereo visionSupports X3D, COLLADA, CityGML
Data Processing for Lidar and Video in MatlabLidarAerial & TerrestrialDeveloped & Improved Matlab / Lidar FunctionsSegmentationGround / Vegetation / BuildingsImplemented parsing and processing algorithmsObject ConstructionSurface ModelVegetation Building Models
VideoDr. Manjunaths project features >29 operational (Oct 2010) cameras across campusOverlay data as video texture in virtual worldParse objects via several mobile georeferenced video feeds?
Surface issuesLevel of detail LOD support is present in X3D browsers & easily implemented for grids.Elevation grids are inefficient. Waste of bits on relatively flat terrain. Example: South campus (1200 x 1500 meters) wont render as a ground surface in Instant PlayerLOD/TINs?X3D interface for object ID corrections?
Building and Vegetation Extraction & IDIssuesQuality of fitted polygonsNumber of surfacesEfficiency in storageSolutionsTry newer methods of object extractionProgressive filter removes vegetation before buildingsData fusionBalance issues of better modeling and automation
Fusion Approaches:Campus Flora Project
Video Attached to Terrestrial LidarHigh density scans at 20 locations on campusOne group on campus working on overlaying video onto these points.Memory requirements? Feasibility for X3D?
Collaboration EnvironmentMulti-user visualizationDistributed AnalysisMany analysts in separate spacesNetworked desktops, ready resourcesCentralized AnalysisMany analysis in a large single spaceAllosphere and other immersive environments
Virtual Reality: The Allosphere
Applications and Test RunsLidar filter error visualization and correctionIndoor-outdoor interactionPlaying out scenarios of a hostage sceneMultiple viewpointsOpenable buildingsIn-scene video:Scene following in immersive 3D via moving textured video (campus tour)Video texturing on groundVisualizing campus traffic gathered from camera networkOverlay video on scene (video texturing)Cartographic representations of trafficParsing dynamic geoObjects from video stream with geolocated (x,y,z,h,p,r,R,G,B) sensorsDynamic sensors (e.g., Android phones)Important to have portals back to raw video stream
AcknowledgementsIC Postdoc Fellowship ProgramIC Advisor: Greg SmithPI: Keith ClarkeResearchers: William McBride
~10km x 1.6km**Source for video game data: http://www.npd.com/press/releases/press_100114.html*