Brad Grinstead - PowerPoint PPT Presentation

Loading...

PPT – Brad Grinstead PowerPoint presentation | free to download - id: 15bb23-MGRhY



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Brad Grinstead

Description:

Brad Grinstead – PowerPoint PPT presentation

Number of Views:80
Avg rating:3.0/5.0
Slides: 21
Provided by: brunopvan
Learn more at: http://imaging.utk.edu
Category:
Tags: brad | fabio | grinstead

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Brad Grinstead


1
Fast Digitization of Real World Environments for
Evaluation of Driver Behavior
  • Brad Grinstead
  • Michael Roy
  • UTK Advisors Mongi Abidi, Andreas Koschan

2
Outline
  • Introduction
  • Motivation
  • Concept
  • Objectives
  • Technical Approach
  • Geometry From Laser Range Scanning
  • Interprofile Pose Estimation
  • Generating 3D Models
  • Multiresolution Processing
  • Results
  • Conclusion
  • Discussion
  • Future Work

3
Why?
http//www.nads-sc.uiowa.edu/
Realism in simulations!
http//www.tacom.army.mil/tardec/nac/index.htm
4
Concept
5
Objectives
Develop a system for the fast digitization of
real environments
  • Utilize vehicle-borne scanning
  • Scanner system
  • Hardware for inter-profile alignment
  • Increased verisimilitude over previous methods
  • Results can be used for a variety of tasks
  • Driver performance evaluation
  • Urban planning
  • Others

6
Outline
  • Introduction
  • Motivation
  • Concept
  • Objectives
  • Technical Approach
  • Geometry From Laser Range Scanning
  • Interprofile Pose Estimation
  • Generating 3D Models
  • Multiresolution Processing
  • Results
  • Conclusion
  • Discussion
  • Future Work

7
Overall Approach
Mobile Range Scanning System
Textured 3D Models
Geometry and Texture Information
Path Determination from Multiple Sensors
8
How?
The range finder electronics (1) send out a pulse
of infrared laser light and keep track of the
amount of time that elapses until the beam
returns, thus determining the distance to the
target object. The laser beam (2) is deflected
by a set of rotating mirrors (3) in order to
acquire the range information for a line of
points. A frame scan is provided by rotating the
optical head (4) through the desired scan angle
(up to 333). The acquired information is
provided via a parallel port connection (5) to a
laptop (6) or PC. The PC converses with the
scanner with the RIEGL 3D-RiSCAN software (7),
which provides basic acquisition, sensor
configuration, visualization, and manipulation
functions.
9
Range Imagery to 3D Models
Target Scene
Pseudocolored Range Image
3D Model from several views
10
Imaging System

Individual profiles are converted to a global
reference frame
R is formed from the roll, pitch, and yaw
parameters of the scanner
11
Pose Estimation
Inertial Measurement Unit
Global Positioning System
12
Multiresolution Processing
  • Level of Detail (LOD) representation allows
    view-dependant triangulation

We can also use multiresolution analysis to
perform intelligent denoising of surface models
13
Outline
  • Introduction
  • Motivation
  • Concept
  • Objectives
  • Technical Approach
  • Geometry From Laser Range Scanning
  • Interprofile Pose Estimation
  • Generating 3D Models
  • Multiresolution Processing
  • Results
  • Conclusion
  • Discussion
  • Future Work

14
Results Local Supermarket
3D geometry
Images of building
Composite texture image
3D Model
Complete model 876,000 triangles Acquired 150 m
worth of data in 10 seconds
15
Results Womens Basketball Hall of Fame
3D geometry
Composite texture image
Image of building
3D Model
Complete model 2.8 million triangles Acquired
600 m worth of data in 32 seconds
16
Outline
  • Introduction
  • Motivation
  • Concept
  • Objectives
  • Technical Approach
  • Overview
  • Geometry From Laser Range Scanning
  • Interprofile Pose Estimation
  • Generating 3D Models
  • Multiresolution Processing
  • Results
  • Conclusion
  • Discussion
  • Future Work

17
Wrap Up
  • A system is under development for the fast
    digitization of environments
  • Geometry and texture have been acquired
  • Pose instrumentation has been integrated
  • The acquired models are true to life in terms of
    geometry, scale, and color
  • These models are appropriate for integrating with
    existing simulators to evaluate driver
    performance in a realistic setting

18
Future Efforts
  • Feature Matching and Tracking from Video
  • Use standard feature matching and tracking from
    video sequences to provide
  • Additional pose information for interprofile
    alignment.
  • 2. 3D information for missing geometries (due to
    occlusions, reflections, etc.)

Consecutive images from a video sequence
19
Future Efforts
Foreground Object Removal and Hole
Filling Remove non-informative small scale
foreground objects and use an intelligent hole
filling method (volumetric diffusion, 3D from
video, etc.) to fill in the surface holes.
Combine this with high-resolution color imagery
for texture information.
Road Scanning for Terrain Analysis Utilize the
same scanning methods to acquire accurate surface
models for terrain modeling and simulation
purposes
20
Thank You
Questions?
http//imaging.utk.edu
About PowerShow.com