Zürich Autonomous Systems Lab Rainer Voigt 28. September 2011.

8
Zürich Autonomous Systems Lab Robust Embedded Egomotion Estimation with Application to Aerial Inspection Vehicles Rainer Voigt 28. September 2011 Name - Short Title

Transcript of Zürich Autonomous Systems Lab Rainer Voigt 28. September 2011.

Page 1: Zürich Autonomous Systems Lab Rainer Voigt 28. September 2011.

ZürichAutonomous Systems Lab

Robust Embedded Egomotion Estimation

with Application to Aerial Inspection Vehicles

Rainer Voigt

28. September 2011

Name - Short Title

Page 2: Zürich Autonomous Systems Lab Rainer Voigt 28. September 2011.

Auto

nom

ous

Syst

em

s La

b

ZürichRainer Voigt – Robust Embedded Egomotion Estimation2

Inspection of e.g. Thermal Power Plants

Boilers up to 100 meters high Up to 1 Mio USD per day of

outage UAV for visual pre-inspection

Introduction | Motion Estimation | Results | Video

AIRobots Innovative Aerial Service for Remote Inspections by Contact

Page 3: Zürich Autonomous Systems Lab Rainer Voigt 28. September 2011.

Auto

nom

ous

Syst

em

s La

b

ZürichRainer Voigt – Robust Embedded Egomotion Estimation3

Introduction | Motion Estimation | Results | Video

Motion Estimation

HDR, Global Shutter

CMOS Sensors

3x Gyroscopes3-Axis

AccelerometerHigh Power LEDs

Front-Looking Stereo-IMU Rig Motion Estimation with Sparse Features Tight Coupling of Visual and Inertial Cues for

Feature Disambiguation on repetitive Texture

Page 4: Zürich Autonomous Systems Lab Rainer Voigt 28. September 2011.

Auto

nom

ous

Syst

em

s La

b

Zürich4

Extract key points Compute descriptors Associate features of

left and right image Triangulate features to

obtain 3D points

Introduction | Motion Estimation | Results | Video

Rainer Voigt – Robust Embedded Egomotion Estimation

Visual Egomotion

Page 5: Zürich Autonomous Systems Lab Rainer Voigt 28. September 2011.

Auto

nom

ous

Syst

em

s La

b

Zürich5

Extract key points Compute descriptors Associate features of

left and right image Triangulate features to

obtain 3D points

Introduction | Motion Estimation | Results | Video

Rainer Voigt – Robust Embedded Egomotion Estimation

Visual Egomotion

Page 6: Zürich Autonomous Systems Lab Rainer Voigt 28. September 2011.

Auto

nom

ous

Syst

em

s La

b

Zürich6

Capture next stereo shot Compute key points,

descriptors and 3D points Predict feature locations

via IMU motion constraints Associate features between

both stereo pairs RANSAC outlier rejection Refinement via

bundle adjustment

Introduction | Motion Estimation | Results | Video

Rainer Voigt – Robust Embedded Egomotion Estimation

Visual Egomotion

Page 7: Zürich Autonomous Systems Lab Rainer Voigt 28. September 2011.

Auto

nom

ous

Syst

em

s La

b

ZürichRainer Voigt – Robust Embedded Egomotion Estimation7

Introduction | Motion Estimation | Results | Video

Results

Robust real-time motion estimation on single core Intel Atom CPU

Vehicle stabilization

Page 8: Zürich Autonomous Systems Lab Rainer Voigt 28. September 2011.

Introduction | Visual Egomotion | Vision Covariance | EKF | Feature Prediction | Results | Outlook

8