17
Internship Presentation Marius Muja August 26, 2009

Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

  • Upload
    others

  • View
    3

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

Internship Presentation

Marius Muja

August 26, 2009

Page 2: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

What’s been keeping me busy?

Milestone 2

Far outlet detectionDoor handle detection

Tabletop object detection

2D Chamfer matching3D Model fitting

Integrating FLANN in OpenCV

PR2 Challenge

Page 3: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

Far outlet detection

Uses the stereo camera andbase laser to identify locationand 3D pose of power outlets

Outlet candidates obtained bysegmenting the disparity image

Wall pose computed form thebase laser scan

List of outlet candidates isfiltered using position, size andorientation priors

Page 4: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

Far outlet detection

Outlet candidate patches arerectified to obtain a frontal view

Outlet identity is confirmedusing a template matchingalgorithm

Page 5: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

Door handle detection

Initial detection using a boostedcascaded classifier of haarfeatures

False positives eliminated usingposition and size priors

Detections clustered acrossseveral frames for increasedrobustness

3D location of handle fromstereo

Page 6: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

Tabletop object detection

Manipulating tabletop objects: one of the main tasks of arobot in a household environment

Setting/cleaning the tableServing drinks

Usual tabletop objects are difficult (no texture, transparent)

Techniques based on local features fail

Page 7: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

Ikea objects dataset

Page 8: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

2D Chamfer matching

Approach for finding the bestmatch of a contour model to anedge image

Contours can be strongindicator of object’s identity

Page 9: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

2D Chamfer matching

Page 10: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

3D model fitting

Grasping of delicate (glass)objects requires precise objectlocalization

Just a bounding box aroundthe object is not enoughWe must also know the grasplocation for a specific object

Extended the 2D chamfermatching approach to 3D

Page 11: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

3D model fitting

Two stage approach

Bottom-up object localization

Determines probable objectlocationsTable plane detection andremovalPoint cloud clustering

Top down model fitting

Determines exact object poseand identityFind the model with the bestcorrespondence to the pointcloudICP-like (Iterative ClosestPoint) algorithm

Page 12: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

3D model fitting

“tabletop objects” package

The 3D model fitter determines

Object identityObject poseGrasp poseObject mesh - used in theplaning stage for a moreprecise collision map

Integrated with the planingpipeline

Page 13: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

PR2 challenge

Applied the 3D model fittingapproach for detectingjuice/watter bottles

Used the tilting laser pointcloud (much more sparse thanthe laser point cloud)

Page 14: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

Model capture

Page 15: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

Others

Integrated FLANN in OpenCV

FLANN - Fast Library for Approximate Nearest Neighborsfast nearest-neighbor searching in high dimensional spaces

PR2 teleoperation using a phone

based on Asterisk open source PBXcall the robot on the phone and send it to a specific officeother use cases possible: eg. deliver a message

Page 16: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

Work in progress

Improving bottom-up part of object detection pipeline

3D features voting for object pose

3D object tracking

Neigborhood indexing for large point cloud structures

Sparse voxel grid indexingBenchmarks for the different index types

Page 17: Internship Presentation - ROS.org...Internship Presentation Marius Muja August 26, 2009 What’s been keeping me busy? Milestone 2 Far outlet detection Door handle detection Tabletop

Thank you!