Robotics Project Update

1
Robotics Project Update

• Brian Clipp
• November 16, 2006

2
3D Simultaneous Localization and Mapping of Urban
Scenes
3
3D Simultaneous Localization and Mapping of Urban
Scenes

• Data from hand-held camera and GPS unit
• Video GPS - Scene Geometry Camera Poses
• Use Kalman Filter and Smooth motion model to
  estimate camera motion
• GPS allows recovery of model scale and position
  in the world

4
Drift In Vision Only Estimation
5
Example Video and GPS
6
3D SLAM Using Video and GPS Measurements

• Progress So Far
• Data for testing acquired from Marc
• Wrote AVI reader into Urbanscape pipeline to
  allow reading data.
• Started work on GPX (GPS file format) reading
  code.
• Developed Kalman Filter tracker using inertial
  and GPS measurements

7
To Do

• Finish GPX reading ( 2hr)
• Extend Existing KF to work with GPS only ( 1
  month)
• Eliminate grossly wrong GPS measurements (due to
  multipath for example) before processing.
• Generate estimates of initial camera velocity and
  position.
• Estimate vertical vector in the scene
• Needed to recover camera orientation
• RANSAC to find large intersecting planes in
  sparse 3D points
• Write camera positions and features to KML for
  display in Google Earth (unlikely but if time
  permits)