V I S I O N L A B S GPU-Powered Megacity Scale Transport Management, - - PowerPoint PPT Presentation

▶
v i s i o n l a b s
SMART_READER_LITE
LIVE PREVIEW

V I S I O N L A B S GPU-Powered Megacity Scale Transport Management, - - PowerPoint PPT Presentation

Dec 08, 2022 25 likes •218 views

V I S I O N L A B S GPU-Powered Megacity Scale Transport Management, Municipal Services and Public Safety Solutions Talk ID: S8584 | Anton Nazarkin, Global BDO VisionLabs About: Customers: We are proud to serve the industry leaders and SMBs

slide-1
SLIDE 1

V I S I O N L A B S

GPU-Powered Megacity Scale Transport Management, Municipal Services and Public Safety Solutions Talk ID: S8584 | Anton Nazarkin, Global BDO

slide-2
SLIDE 2

02 VisionLabs is a team of Computer Vision and Deep Learning

  • experts. Founded in 2012, today we offer face recognition

products and custom Computer Vision solutions globally

We are proud to serve the industry leaders and SMBs

The flagship products of VisionLabs are LUNA SDK facial analysis and recognition engine and LUNA PLATFORM biometric data management system. We also provide solutions for AR/VR field (3D face avatars) and Smart City field (traffic analysis and classification).

We are happy to work together with the global technology leaders

About: Customers: Products: Collaboration: VisionLabs

slide-3
SLIDE 3

03

PyTorch contributor

Companies & Universities developing PyTorch

slide-4
SLIDE 4

04

Outstanding facts

Our production face recognition algorithm was independently tested by NIST and is TOP rated for accuracy and speed We are the only company in the TOP 5 according to the LFW test result having commercially available algorithm with the same performance characteristics More than 1 000 000 cameras are streaming video to our products daily with the largest single Customer implementtion of 250 000+ cameras Our products process more than 60B face recognition requests per year globally with the largest single Customer implementation of 15B requests per year

slide-5
SLIDE 5

05

Accuracy

  • TPR at FPR 10-3 ≈ 99,18%
  • TPR at FPR 10-6 ≈ 99,00%
  • TPR at FPR 10-8 ≈ 92,44%

TPR: True Positive Rate FPR: False Positive Rate

Face Descriptor Size ~ 256 bytes

slide-6
SLIDE 6

06

Deep learning dataset

10M+ face images

  • Access to data from multiple device types, face

capture scenarios with varying quality settings and environment conditions guarantees robust algorithm operation

  • Intense use of NVIDIA GPUs for training;

parallel training on 4-8 GPSs gives faster training and better accuracy with larger batch sizes

slide-7
SLIDE 7

07

Privacy

Customers’ data is 100% encrypted and 100% protected against unauthorized access

Handling of personal data is fully aligned with local laws and corporate IT security

slide-8
SLIDE 8

14

VL plug-in for NVIDIA DriveIX

  • VL Driver Monitoring

Integrated as a plug-in for DriveIX

  • Jetson or DrivePX box

solution for public transport security

  • Real-time face-based

immobilizer and driver state monitoring

slide-9
SLIDE 9

09

Driver monitoring (demo)

slide-10
SLIDE 10

11

  • Takeover prevention

in case of terrorists attacks

  • Not the appointed

driver = engine stop

  • Driving time and

work hours control

Public transport, service transport, car sharing

slide-11
SLIDE 11

15

Megacity traffic monitoring

NVIDIA TESLA P40

Megacity scale vehicle traffic analysis and anomalies detection powered by NVIDIA Tesla P40 with 80M+ recognition requests daily

slide-12
SLIDE 12

08

Traffic monitoring (demo)

slide-13
SLIDE 13

08

Traffic monitoring (demo)

slide-14
SLIDE 14

21

National scale biometric platform

NVIDIA TESLA P40

National scale face identification platform for financial services with 110M+ faces database.

slide-15
SLIDE 15

22

Cross-platform face recognition

  • Set of front-end modules for

face detection, best shot selection and normalization

  • Stable cross-platform
  • peration and simple

configuration for multiple architectures

  • Ultra-fast GPU powered

backend face recognition platform with extremely low latency

ATM / Terminal WEB-browsers Mobile devices IP-cameras Personal computers Door bell Server

slide-16
SLIDE 16

27

ID Management

slide-17
SLIDE 17

23

  • Visitor counting and Customer

dwell time estimation

  • Accurate gender (99.5%) and

age (+- 1,5 years) detection

  • Detailed marketing research

based on impersonal data analysis

Audience Measurement

slide-18
SLIDE 18

701 679 845 897 854 796 872 796 635 331 264 332 399 249 359 281 163 255 100 82 209 275 96 157 332 409 118 358 207 120 117 286 168 195 321 386 239 350 406 359 410 359 409 174 191 100 200 300 400 500 600 700 800 900 1000

Quantity

14.08.2017 15.08.2017 15.08.2017 16.08.2017 17.08.2017 18.08.2017 19.08.2017 20.08.2017 21.08.2017

Visitors distribution by zone

25

Alcohol cashbox Enter Cashbox_1 Cashbox_2 Cashbox_3

slide-19
SLIDE 19

GBDO, VisionLabs

E-mail: sales@visionlabs.ai Web: www.visionlabs.ai

Anton Nazarkin