Daniel Munoz: Indoor Object Detection

Willow Garage Summer Finale Presentation Dan Munoz

Mentor: Kurt Konolige

CMU Research Geometry Surface Estimation (Hoiem et al.)

Building

Sky

Vertical Support

3-D Point Cloud Classification

Vegetation Ground Tree trunk

Contextual Classification with Functional Max-Margin Markov Networks D. Munoz, J. A. Bagnell, N. Vandapel, M. Hebert CVPR 2009

2

Outdoor Classification

Vegetation

Tree-trunk

Ground

Building 3

Summer 09

Floor

Ceiling

Column

Chair

Table

Wall 4

Challenges  Training

data

 Discriminative

features

 Learning

5

Training data  Issues

with getting 3-D labels from 2-D images

 Manually

created labeled dataset of room-sized objects (chairs, tables, trash cans, etc.) in PCD format 6

Discriminative Features  descriptors_3d

• Utilizes PCML routines • Similar interface with image descriptors_2d (Alex T.) • Parallelized with OpenMP*  Example

features:

Image from Johnson and Hebert 1999

 Works

over groups/clusters of points as well

• point_cloud_clustering (k-means, single-link, local neighbors)

7

Learning  functional_m3n

• ROS agnostic • Can do online learning • Extended implementation of CVPR’09 work 1

min tutorial…

8

Independent Classification

9

Local Interactions

10

Using Higher Order Information

Colored by elevation 11

Region Based Model

12

Simple Algorithm  For

T iterations

• Classify with current model

• Create training set D from misclassifications  (Over features from each clique)

D

• Train favorite classifier  OpenCV regression trees

• Add classifier to model 13

Experiments

14

Table-top Objects

15

Table-top Objects  Only

3-D features (worst example)

Stapler

Mouse

Mug

Background 16

Adding Image Features

17

Results

Stapler

Mouse

Mug

Background 18

Results

Stapler

Mouse

Mug

Background 19

Results

Stapler

Mouse

Mug

Background 20

Results

Stapler

Mouse

Mug

Background 21

Results

Stapler

Mouse

Mug

Background 22

Results

Stapler

Mouse

Mug

Background 23

Room-sized Objects

Floor

Ceiling

Column

Chair

Table

Wall

Cabinets 24

Room-sized Objects

Floor

Ceiling

Column

Chair

Table

Wall

Cabinets 25

Room-sized Objects

Floor

Ceiling

Column

Chair

Table

Wall

Cabinets 26

More experiments required  There

were no cabinets off the ground in training set

Floor

Ceiling

Column

Chair

Table

Wall

Cabinets 27

Available Future Work  descriptors_3d

• Faster neighborhood data structure (Marius, Ethan) • Point Histogram Features (Radu, Gary) • 3-D Chamfer distances (Marius)  point_cloud_clustering

• Ground plane removal and then clustering 2-D projections (Caroline) • Mean-shift, etc.  functional_m3n

• Random Forests, Neural nets (OpenCV) • Boosted spheres (Alex T.) 28