VideoTrace: Interactive 3d modelling for all Anton van den Hengel Director, Australian Centre for Visual technologies Associate Professor, Adelaide University, South Australia Director, PunchCard Visual technologies
Input
Modelling
Results
Interactive 3D modelling 3D modelling is critical to all sorts of application Special effects, but also mining, architecture, defence, urban planning, … People are getting more visually sophisticated More 3D data is being generated More cameras, but also scanners etc The interfaces of modelling programs are usually very hard to fathom
Why? Insert your own objects into a game Put your couch into second life Model your house for Google Earth Video editing Cut and paste between sequences Remove someone from your home videos
Put your truck into a game
Put your truck into a game
Modelling for animation
Video editing requires models
Dense surface reconstruction
Video editing requires models
Structure from motion
The process Capture and import the video Perform structure and motion analysis Interact with the system to generate and edit the model Export to your application
The approach Pre-compute where possible Then interact Structure from motion (camera tracking) Superpixels Then interact Interactions allow user to exploit precomputed results
Structure from motion Camera tracking Calculates Reconstructed point cloud Camera parameters Location Orientation Intrinsics (eg. Focal length) Informs interaction interpretation process
Interactive modelling from video
Interactions Straight lines Curves Mirroring Extrusion Closed sets of lines define planar polygons Curves For planar shapes with curved edges For NURBS surfaces Mirroring Duplicates existing geometry Extrusion
Fitting planar faces User specifies boundary Boundary specifies infinitely many planes Similar to pre-emptive RANSAC Generate bounded plane hypotheses from point cloud Eliminate hypotheses that fail a series of tests Run simplest / most robust tests first Generally 3d tests before 2d tests
Fitting planar faces Line of sight Object points Image plane
Hierarchical RANSAC Generate bounded plane hypotheses Tests Support from point cloud Reprojects within new image boundaries Constraints on relative edge length and face size Colour histogram matching on faces Colour matching on edge projections Reprojection is not self-occluding
Curves
Mirroring
Extrusion
Dense surface reconstruction Needs to be at interactive speed Calculated as a max-flow graph-cut over a Markov Random Field Link cost based on photoconsitency
Modelling without features
Modelling without features
Recent model
Future work Other interactions Other data sources Occluding contours Interactive SFM De / Re-lighting