robocup vision tracking with xetal processor

24
Philips Research - -1 PR-AvG-1 Robocup Vision Tracking with Xetal Processor Edge and colour-based object detection Sebastien Pierrot Supervisors: Harry Broers (CFT), Anteneh Abbo, Richard Kleihorst (NATLAB)

Upload: wilkinson-junior

Post on 30-Dec-2015

34 views

Category:

Documents


0 download

DESCRIPTION

Robocup Vision Tracking with Xetal Processor. Edge and colour-based object detection. Sebastien Pierrot. Supervisors: Harry Broers (CFT), Anteneh Abbo, Richard Kleihorst (NATLAB). Outline. Introduction Vision System Object Tracking Future Work. Introduction (1). - PowerPoint PPT Presentation

TRANSCRIPT

Philips Research

- -1

PR-AvG-1

Robocup Vision Tracking with Xetal Processor

Edge and colour-based object detection

Sebastien Pierrot

Supervisors: Harry Broers (CFT),

Anteneh Abbo,

Richard Kleihorst (NATLAB)

Philips Research

- -2

PR-AvG-2

Outline

• Introduction

• Vision System

• Object Tracking

• Future Work

Philips Research

- -3

PR-AvG-3

Introduction (1)

Object Colour Ball Orange Field and Corners Green Goals Blue and

Yellow Walls and Lines White Robots and Logos on walls

Black

Team Shirts Magenta and Cyan

Philips Research

- -4

PR-AvG-4

Machine Vision

Blobs Analysis

Capture

ObjectDetection

Compression

Communication

Pixel to worldtranslation

Introduction (2)

Philips Research

- -5

PR-AvG-5

Outline

• Introduction• Vision System

– Robocup vision system– Xetal Architecture– Task division

• Object Tracking • Future Work

Philips Research

- -6

PR-AvG-6

Actual Robocup vision systemColour processing

colour camera FugaRGB-YUV box

High-Speed Monochrome Processing

B/W camera Fuga

Vision system (1)

Robocup vision system evolution

Xetal processor

Color MOS

Sensor

Trimedia Digital I/O

Xetal processor

B/W CMOS Sensor

Trimedia Digital I/O

Philips Research

- -7

PR-AvG-7

Xetal Architecture

Block Schema

Vision system (2)

Philips Research

- -8

PR-AvG-8

TRIMEDIA CHIPTRIMEDIA CHIP

XETALCHIPXETALCHIP

Camera

Repartition Tasks

Blobs Analysis

Capture

ObjectDetection

Compression

Communication

Pixels to worldtranslation

Communication

Communication

Vision system (3)

Philips Research

- -9

PR-AvG-9

Outline

• Introduction• System Vision• Object Tracking

– Color-based detection– Edge detection

• Future Work

Philips Research

- -10

PR-AvG-10

Color-based detection (1)

RGB

(0,0,0)

(0,1,0)

(1,0,0)

Green

Black

Blue(0,0,1)

Yellow

Cyan

White(1,1,1)

Magenta

3-D RGB cube

Object Tracking (1)

Philips Research

- -11

PR-AvG-11

YUV color space

Y : Luminosity

U,V: Chromatic components

Y=0.3*R+0.58*G+0.12*B

U=0.17*R-0.33*G+0.5*B

V=0.5*R-0.42*G+0.08*B

Color-based detection (2)Object Tracking (2)

Philips Research

- -12

PR-AvG-12

HSV color space

V = ( R + G + B )/3

S = ( 1 - min(R,G,B)/ V )

H = 0 + (G-B)/ if max is R

= 1/3 + (B-R)/ if max is G

= 2/3 + (R-G)/ if max is B

is the (max-min) of the RGBs

Color-based detection (3)Object Tracking (3)

V : Value

S :Saturation

H : Hue

Philips Research

- -13

PR-AvG-13

HSI color space

=/2 if G>B=3/2 if G<BH=1 if G=B

Blue

I

Cyan

Red

GreenYellow

Magenta

Black

S

H

3

2

2

320.5

3arctanα2π1

GBRI

BGBGRS

BGIRH

Color-based detection (4)Object Tracking (4)

I : Intensity

S :Saturation

H : Hue

Philips Research

- -14

PR-AvG-14

Segmentation examples

U

V

S

H

Linear

Constant

Color-based detection (5)Object Tracking (5)

Philips Research

- -15

PR-AvG-15

Object Tracking (6)Color-based detection (6)

Orange YUV segmentation

Philips Research

- -16

PR-AvG-16

Object Tracking (7)Color-based detection (7)

Orange HSI segmentation

Philips Research

- -17

PR-AvG-17

Implementation discussion• HSV

4 variable divisions• HSI

One variable divisionArc tangent function

Conclusion: • YUV linear segmentation for quicker processing• HSI constant segmentation for tuning facility and better

color density

Color-based detection (8)Object Tracking (8)

Philips Research

- -18

PR-AvG-18

Goal• Strong intensity contrast detection

• Divide the image into areas

corresponding to different objects

• Reducing image informations

ComputationEstimated with the maximum of the 1st derivative or with

the zero crossing of the 2nd derivative

Edge detection (1)Object Tracking (9)

Philips Research

- -19

PR-AvG-19

Sobel edge detectorApproximation absolute gradient magnitude at each point in an input grayscale image

a pair of 3×3 convolution kernels

• Advantage: Simple implementation

• Drawback: Sensible to the noise

Edge detection (2)Object Tracking (10)

Philips Research

- -20

PR-AvG-20

Canny edge detectorMore sophisticated: multi-stage process

• Advantages– Simple thresholing

– Lower sensibility to the noise

– Large patterns: 5*5,7*7…

• Drawbacks– Larger code program

Edge detection (3)Object Tracking (11)

Philips Research

- -21

PR-AvG-21

G’’(2)G’(0)

PE index -2 -1 0 1 2 3

Line Mem0 a11 a12 a13 a14 a15 a16 a17

Line Mem1 a21 a22 a23 a24 a25 a26 a27

Line Mem2 …..

Line Mem3 …..

Line Mem4 …..

Line Mem5 a61 a62 a63 a64 a65 a66 a67

Line Mem6 a71 a72 a73 a74 a75 a76 a77

7*7 Kernel

Example: 7*7 pattern elaboration

• Shifts

• Sum of intermediate

Results:

Gx/y(0)= G’(0)+G’’(2)

Gx/y(1)= G’(1)+G’’(3)

Edge detection (4)Object Tracking (12)

Philips Research

- -22

PR-AvG-22

Canny detector

Sobel detector

Edge detection (5)Object Tracking (13)

Results

Philips Research

- -23

PR-AvG-23

Outline

• Introduction• System Vision• Object Tracking • Future Work

– Edge detection tuning– Data compression

Philips Research

- -24

PR-AvG-24

Compression

• Goal: Reducing space information

• Proposed format:

• Features– Processor identification (PID)

– Statistic information delivering from serial processor

– Shifts for blank elimination

Min1 Color1 Max1 Min2 Color2 Max2 0000