Crowdsourcing 3D Semantic Maps for Vehicle Cognition Cognition for - - PowerPoint PPT Presentation

▶
crowdsourcing 3d semantic maps for vehicle cognition
SMART_READER_LITE
LIVE PREVIEW

Crowdsourcing 3D Semantic Maps for Vehicle Cognition Cognition for - - PowerPoint PPT Presentation

Oct 24, 2022 273 likes •574 views

Crowdsourcing 3D Semantic Maps for Vehicle Cognition Cognition for Cars Decisions Eyes Cognition Civil Maps Cloud Bounding boxes Civil Maps In Car Vehicle Cognition Vehicle Cognition through 3D Maps Vehicle Cognition through 3D Maps

slide-1
SLIDE 1

Crowdsourcing 3D Semantic Maps for Vehicle Cognition

slide-2
SLIDE 2

Cognition Eyes Civil Maps Cloud Civil Maps In Car

Bounding boxes

Decisions

Cognition for Cars

slide-3
SLIDE 3

Vehicle Cognition

slide-4
SLIDE 4

Vehicle Cognition through 3D Maps

slide-5
SLIDE 5

Vehicle Cognition through 3D Maps

slide-6
SLIDE 6

Vehicle Cognition through 3D Maps

slide-7
SLIDE 7

Vehicle Cognition through 3D Maps

slide-8
SLIDE 8

Vehicle Cognition through 3D Maps

slide-9
SLIDE 9

Vehicle Cognition through 3D Maps

slide-10
SLIDE 10

Vehicle Cognition through 3D Maps

slide-11
SLIDE 11

Vehicle Cognition through 3D Maps

slide-12
SLIDE 12

Vehicle Cognition through 3D Maps

slide-13
SLIDE 13

3D Semantic Maps

slide-14
SLIDE 14

3D Semantic Maps

slide-15
SLIDE 15

3D Semantic Maps to Cognition

Maps on Earth Maps Projected into a Field of View Maps Projected onto a Camera View (requires precise localizaiton)

slide-16
SLIDE 16

6D Localization

slide-17
SLIDE 17

Sensor data -> 3D Maps & Localization

Sensors Point Cloud 3D Semantic Map Localization Map

slide-18
SLIDE 18

Problem: GPS is not reliable or accurate

  • Actual vehicle GPS traces collected in downtown San Francisco
slide-19
SLIDE 19
slide-20
SLIDE 20

Localization Methods: ICP VS Signatures

SLOW: ICP FAST: Signatures

  • Point to point comparisons
  • 2 million points per second
  • Lots of CPU / GPU usage
  • Signature to Signature comparisons
  • 1000 Signature per second
  • ARM Cortex is enough
slide-21
SLIDE 21

Synthetics : AI Training Process

Synthetic Sensor Data AI Models 3D Semantic Maps Real Sensor Data Procedural 3D modeling Real World Driving Training 90% Validation 10%

slide-22
SLIDE 22

Crowdsourced 3D Semantic Maps

slide-23
SLIDE 23

Crowdsourced Maps: Client/Server

... ...

Local Map Cache Map Auditing Aggregation (Completeness) Averaging, Outlier rejection

  • Hyp. Testing,

Job Dispatching Semantic Derivation

slide-24
SLIDE 24

Crowdsourced Maps: Supply & Demand

Map Contributions Map Usage Civil Maps

Map Aggregation

slide-25
SLIDE 25

Civil Maps Github: videos & code ○ https://civilmaps.github.io/cm-hal/ Session Three: (Sensor Fusion Part 1): Thursday, May 25th 1:30 PM PDT (4:30 PM EST) Session Four: (Sensor Fusion Part 2): Thursday, June 29th 1:30 PM PDT (4:30 PM EST) Getting Started With Sensor Data

Free Webinar Series by Civil Maps (4 Sessions)

slide-26
SLIDE 26

Towards Mixed Reality Programmable Roads

slide-27
SLIDE 27

Engineering Blog on Medium: https://medium.com/@CivilMaps Stay in Touch:

slide-28
SLIDE 28

Cognition for Cars

Contact

Fabien Chraim, VP Research and Development fabien@civilmaps.com Scott Harvey, Sr. Machine Vision Engineer scott@civilmaps.com