from 3d scene geometry to human workspace
DESCRIPTION
From 3D Scene Geometry to Human Workspace. Authors: Abhinav Gupta, Scott Satkin , Alexei A. Efros and Martial Hebert The Robotics Institute, Carnegie Mellon University Presenter: Jiapei Zhang. Project Undertaking. Image if I were Abhinav - PowerPoint PPT PresentationTRANSCRIPT
From 3D Scene Geometry to Human Workspace
Authors: Abhinav Gupta, Scott Satkin, Alexei A. Efros and Martial Hebert
The Robotics Institute, Carnegie Mellon UniversityPresenter: Jiapei Zhang
Project Undertaking
Image if I were Abhinav
I received a project from my boss about indoor scene understanding
Selection of Subject Material at hand
• A set of indoor-scene images• Motion Capture Database• No LiDAR scanner, not as rich as VCC, China• How I wish to be a VCCer one day
VCC V5
Selection of Project Inspirations from Psychology
Where is the “entrance” of my novelty? Where is the “Observer” for traditional Image Understanding? Put Human back to Image Gestalt Psychologists: Objects are best defined not by their
identity but rather by their function. The meaning or a value of a thingconsist of what it affords. -----JJ Gibson(1979) “Affordance” “functionality”
Selection of Project Shoulder of Giants
Bounding Box Level == Object DetectionSingle-View Indoor Geometry EstimationHuman Pose Analysis via Motion Capture
Selection of Project Give a Final Word
Propose a scene representation that combines 3D scene geometry with a set of possible human actions, to create a joint space of human-scene interactions.
Predict the full human workspace of the scene
3D Scene Understanding
HUMAN ACTION MODELING
Implementation Detail Representing Scene Geometry
An indoor scene is modeled by the layout of room(walls, floors, ceilings) and the location/layout of objects.
The objects are modeled by a set of occupied voxels.
Implementation of Detail Geometry Estimation
Sliding Cuboids
Selection
Implementation of Detail Qualitative Human Pose Representation
Each pose is represented by the occupied blocks in discretized 3D space(red) and the required surfaces of interaction(green).
Implementation of Detail Human-Scene Interactions
Constraints
Binary Occupancy Matrix “Space Search” to fit “Constrains”
Free space constrain
Support constrain
Harvest Season Result Showing
Retrospect and Meditation
• Now return back to me• What useful info. for our Indoor-Scene Project
I could extract from this paper ?
Retrospect and Meditation Incorporate images to our projects Will geometry estimation improve our rough mesh model?
Categorize “mobility” of scene into two:a) Extract from Repetitions: Based on Geometry b) Extract from “Human Pose”: Based on Functionality