Multiview Photogrammetry 3D Virtual Geology for everyone A short course Marko Vrabec University of Ljubljana, Department of Geology
FIRST: some background info Precarious structural measurements of fractures in Quaternary slope breccia, Idrija fault zone, Julian Alps
SECOND: course outline 1. A (very) brief introduction to photogrammetry 2. Image acquisition techniques the full rundown coffee break & practical image acquisition 3. Hands-on introduction to the modeling workflow with Agisoft Photoscan software coffee break 4. Special topics: geopositioning of models, hardware considerations, exporting models,... 5. A few selected examples POST-WORKSHOP BARBECUE
HOW DO WE RECORD AND VISUALIZE FIELD INFORMATION? When recording field observations... How successful are we? How efficient are we? How (un)biased are we? How repeateable are our observations?
HOW DO WE RECORD AND VISUALIZE FIELD INFORMATION? When recording field observations... How successful are we? How efficient are we? How (un)biased are we? How repeateable are our observations? 3D virtual geology to the rescue!
3D VIRTUAL GEOLOGY ROCKET SCIENCE OR...? What is the amount of capital and expertise required to enter the virtual 3D world?
3D VIRTUAL GEOLOGY ROCKET SCIENCE OR...? What is the amount of capital and expertise required to enter the virtual 3D world? Democratisation of access to 3D data acquisition and analysis techniques!
PHOTOGRAMMETRY IS......the art, science, and technology of obtaining precise mathematical measurements and three-dimensional (3D) data from two or more photographs (Matthews 2008). The main component necessary for a photogrammetric project is a series of overlapping stereoscopic* images. (*i.e., images showing the same features from varying point(s) of view)
STEREOPHOTOGRAMMETRY everybody is doing it
STEREOPHOTOGRAMMETRY everybody is doing it
METRIC STEREOPHOTOGRAMMETRY NOT everybody is doing it with (relatively) simple triangulation it is possible to calculate 3D positions (x,y,z) of object points from stereo pairs HOWEVER, we need to know either: camera 3D location and pose (yaw, pitch, roll) or 3D coordinates of a series of ground control points (GCPs) appearing in the scene PLUS we want to have as undistorted and metrically correct source images as possible providing this is complicated and expensive
METRIC STEREOPHOTOGRAMMETRY NOT everybody is doing it Traditional equipment requirements: specially configured and calibrated metric cameras (frightfuly expensive) photogrammetric workstation and software plus extensive training and experience
STRUCTURE-FROM-MOTION (SfM) ALGORITHM derived from 1990 s research in computer vision applications a large number of closely spaced, overlapping images ( Multi-view photogrammetry ) camera pose and scene geometry are solved simultaneously results in a 3D point cloud in relative image space (which can be georeferenced later) requires no special equipment, no calibration, no targets, a fully automated procedure a true, revolutionary democratisation of photogrammetry (here we go again...)
STRUCTURE-FROM-MOTION (SfM) ALGORITHM HOW DOES IT WORK? 1) IMAGE ACQUISITION systematically cover your scene with a dense set of highy-overlapping photos (min. 60% overlap), taken from different locations in order to be reconstructed, a single feature must be visible in at least 3 photographs (but the more, the better)
STRUCTURE-FROM-MOTION (SfM) ALGORITHM HOW DOES IT WORK? 2) KEYPOINT EXTRACTION software identifies identical points on different photographs (independent of scale, orientation, illumination, etc.) sensitive to image resolution, sharpness, density also sensitive to object properties and external conditions: material, texture, lighting
STRUCTURE-FROM-MOTION (SfM) ALGORITHM HOW DOES IT WORK? 3) CAMERA POSE ESTIMATION AND SPARSE CLOUD EXTRACTION fully automated sparse bundle adjustment algorithm determination of 3D camera positions and image orientations low-density point cloud extraction
STRUCTURE-FROM-MOTION (SfM) ALGORITHM HOW DOES IT WORK? 3) CAMERA POSE ESTIMATION AND SPARSE CLOUD EXTRACTION fully automated sparse bundle adjustment algorithm determination of 3D camera positions and image orientations low-density point cloud extraction
STRUCTURE-FROM-MOTION (SfM) ALGORITHM HOW DOES IT WORK? 4) DENSE POINT CLOUD EXTRACTION using information from the previous processing step, a dense point cloud is extracted from image data
STRUCTURE-FROM-MOTION (SfM) ALGORITHM HOW DOES IT WORK? 5) SURFACE MODEL CREATION a continuous surface model is generated by triangulation from dense point cloud
STRUCTURE-FROM-MOTION (SfM) ALGORITHM HOW DOES IT WORK? 6) SURFACE TEXTURE CREATION a photographic texture is generated and projected onto the surface model