• No results found

Semantic labelling of pole-like street furniture in mobile laser scanning data : abstract

N/A
N/A
Protected

Academic year: 2021

Share "Semantic labelling of pole-like street furniture in mobile laser scanning data : abstract"

Copied!
22
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Pole-like road furniture interpretation in Mobile Laser

Scanning data

FASHUAI LI

Supervisor: Sander Oude Elberink

Promotor: George Vosselman

(2)

INTRODUCTION

HOW TO CLASSIFY THIS ROAD FURNITURE? MULTI-CLASSES

The interpreted road furniture (Orange: Street signs, Yellow: Street lights, Cyan: Traffic lights, Green: vertical poles, Blue:

Horizontal poles)

Traffic light?

Street light?

Street sign?

(3)

TO SEPARATE ATTACHMENTS

POLE-LIKE ROAD FURNITUE DECMOPOSITION

Road furniture point cloud 2D point density based method RANSAC line fitting method Slice cutting based method

Remove poles and connected components

analysis Precognition

Pole-extraction optimization

(4)
(5)

Street light Traffic sign Street sign

Traffic information sign Traffic light

POLE-LIKE ROAD FURNITURE INTERPRETATION

(6)

 Relative position: top, middle, bottom

 Relative height: from the bottom of the connected pole  Geometric structure: linear, planar, scattered

 Ratio of high reflectance

 Relative angle: the normal of attachments and principal direction of pole  Size: the area

 Ratio of height to length

TO DISTINGUISH DIFFERENT ATTACHMENTS

(7)

 Street lights connected to a vertical pole Top (normally)

Relative height

POLE-LIKE ROAD FURNITURE INTERPRETATION

(8)

 Traffic signs connected to a vertical pole Planar

Relative angle -> perpendicular

Size ->small

Ratio of height to length -> close to 1

TEMPLATES

(9)

 Street signs connected to a vertical pole Planar or linear

Relative angle -> perpendicular

Size -> small

Ratio of height to length ->small

TEMPLATES

(10)

 Traffic information signs connected to a vertical pole Planar

Relative angle -> perpendicular

Size -> large

Ratio of height to length

TEMPLATES

(11)

 Traffic lights connected to a vertical pole

Scattered

Relative height

TEMPLATES

(12)

 Traffic lights connected to a horizontal pole

Connected to horizontal poles

Scattered

TEMPLATES

(13)

 Street signs connected to a horizontal pole Planar or linear

Relative angle -> perpendicular

Size -> small

Ratio of height to length ->small

TEMPLATES

(14)

 Traffic information signs connected to a horizontal pole Planar

Relative angle -> perpendicular

Size -> large

Ratio of height to length

TEMPLATES

(15)
(16)

The evaluation of road furniture interpretation in Enschede dataset

EVALUATION

Visual interpretation FP (%) P S1 S2 T1 T2 T3 Other Total P 111 0 0 0 0 0 0 111 0 S1 1 50 0 3 5 1 5 65 23.1 S2 0 0 7 5 0 0 0 12 41.7 T1 0 1 5 27 1 1 1 36 25.0 T2 0 0 1 0 3 0 0 4 25.0 T3 0 0 0 4 0 16 1 21 23.8 M 7 2 4 14 2 13

(17)

EXPERIMENTAL RESULT

(18)

EXPERIMENTAL RESULT

(19)
(20)

CONCLUSION AND FUTURE WORK

Conclusion

 Fast and better interpretation (more detailed)

 Templates not generic enough (more classes, more templates need to be defined)

Future work

(21)

SWIRLING OF THE RESULT

(22)

Thank you for your attention!

Questions?

Referenties

GERELATEERDE DOCUMENTEN

gegen die Veröffentlichung meines Bildes, meines Wohnhauses, meines Eigentums oder meines Pkw. Man schaue sich mal das Widerspruchs- schreiben an. Viel zu kompliziert. Warum dreht

For the street canyon with tree planting of high crown porosity, increases in concentrations at the leeward wall A and decreases at the windward wall B (Fig. 3, middle) are found

Tot slot maakt de gemeente dankbaar gebruik van de infor‐ matie die Street Pastors leveren over hot spots waar mensen wildplassen of afval laten slingeren, zodat dat er gericht

The whole image predictions of direct prediction from pre-trained FCN model, fine-tuning and FCN model with focal loss function are shown in Figure 5.1 and Figure 5.2 with

It is this fundamental link between no-arbitrage for the price process and its martingale property (15) which lies at the heart of the importance of modern stochastic calculus

A meta-analysis could not be completed for this SR for the following reasons: Some of the outcomes for the SR (for example survival of intervention) were not reported by all

The first sub-model is the privacy calculus model measuring the benefits and risks of information disclosure on the perceived value of disclosure (H1 & H2), the

The night/day accident ratio for different classes of road and light levels (Amsterdam-West and Leeuwarden).. Classification of