AI in Air Traffic Management Christian Thurow, Head of R&D at - - PowerPoint PPT Presentation

▶
ai in air traffic management
SMART_READER_LITE
LIVE PREVIEW

AI in Air Traffic Management Christian Thurow, Head of R&D at - - PowerPoint PPT Presentation

May 12, 2023 180 likes •696 views

AI in Air Traffic Management Christian Thurow, Head of R&D at Searidge WWW.SEARIDGETECH.COM/AIMEE Motivation 1/3 What is Air Traffic Control? 2 Motivation 2/3 Work Increase Annual Growth: 6-7% both #passengers and #flights

slide-1
SLIDE 1

AI in Air Traffic Management

Christian Thurow, Head of R&D at Searidge

WWW.SEARIDGETECH.COM/AIMEE

slide-2
SLIDE 2

2

What is Air Traffic Control? Motivation 1/3

slide-3
SLIDE 3

3

Work Increase

Source: International Civil Aviation Organization, Civil Aviation Statistics of the World and ICAO staff estimates.

  • Annual Growth: 6-7%
  • both #passengers and #flights
  • 2016: 3.7b passengers worldwide

Motivation 2/3

slide-4
SLIDE 4

4

Motivation 3/3

slide-5
SLIDE 5

5

Our Goals

slide-6
SLIDE 6

5

Our Goals

  • reduce controller workload
slide-7
SLIDE 7

5

Our Goals

  • reduce controller workload
  • increase situational awareness
slide-8
SLIDE 8

5

Our Goals

  • reduce controller workload
  • increase situational awareness
  • declutter workspace
slide-9
SLIDE 9

5

Our Goals

  • reduce controller workload
  • increase situational awareness
  • declutter workspace
  • provide additional surveillance data source (added safety)
slide-10
SLIDE 10

6

What does Searidge do?

slide-11
SLIDE 11

6

What does Searidge do?

slide-12
SLIDE 12

7

slide-13
SLIDE 13

8

Challenges

slide-14
SLIDE 14

8

Challenges

  • Building the NN
slide-15
SLIDE 15

8

Challenges

  • Building the NN
  • Training
slide-16
SLIDE 16

8

Challenges

  • Building the NN
  • Training
  • Inferencing Speed
slide-17
SLIDE 17

8

Challenges

  • Building the NN
  • Training
  • Inferencing Speed
  • Safety & Acceptance
slide-18
SLIDE 18

9

  • 1. Challenge: building the NN
slide-19
SLIDE 19
  • company policy: c++

9

  • 1. Challenge: building the NN
slide-20
SLIDE 20
  • company policy: c++
  • first tried caffe, stayed with it

9

  • 1. Challenge: building the NN
slide-21
SLIDE 21
  • company policy: c++
  • first tried caffe, stayed with it
  • first try with VGG16

9

  • 1. Challenge: building the NN
slide-22
SLIDE 22
  • company policy: c++
  • first tried caffe, stayed with it
  • first try with VGG16
  • now VGG19 with custom layers for tracking (37 total)

9

  • 1. Challenge: building the NN
slide-23
SLIDE 23
  • company policy: c++
  • first tried caffe, stayed with it
  • first try with VGG16
  • now VGG19 with custom layers for tracking (37 total)
  • superior performance over previous algorithm

9

  • 1. Challenge: building the NN
slide-24
SLIDE 24
  • company policy: c++
  • first tried caffe, stayed with it
  • first try with VGG16
  • now VGG19 with custom layers for tracking (37 total)
  • superior performance over previous algorithm
  • problems: small objects

9

  • 1. Challenge: building the NN
slide-25
SLIDE 25
  • 2. Challenge: Training

10

slide-26
SLIDE 26
  • 2. Challenge: Training
  • Broad vs. Random Training Initialization?

10

slide-27
SLIDE 27
  • 2. Challenge: Training
  • Broad vs. Random Training Initialization?

10

slide-28
SLIDE 28
  • 2. Challenge: Training
  • Broad vs. Random Training Initialization?
  • How many annotations needed per site?

10

slide-29
SLIDE 29
  • 2. Challenge: Training
  • Broad vs. Random Training Initialization?
  • How many annotations needed per site?
  • Same training set for all airports or specific?

10

slide-30
SLIDE 30
  • 2. Challenge: Training
  • Broad vs. Random Training Initialization?
  • How many annotations needed per site?
  • Same training set for all airports or specific?
  • How many neurons, layers?

10

slide-31
SLIDE 31
  • 2. Challenge: Training
  • Broad vs. Random Training Initialization?
  • How many annotations needed per site?
  • Same training set for all airports or specific?
  • How many neurons, layers?

10

slide-32
SLIDE 32
  • 2. Challenge: Training
  • Broad vs. Random Training Initialization?
  • How many annotations needed per site?
  • Same training set for all airports or specific?
  • How many neurons, layers?
  • How many Epochs?

10

slide-33
SLIDE 33
  • 2. Challenge: Training
  • Broad vs. Random Training Initialization?
  • How many annotations needed per site?
  • Same training set for all airports or specific?
  • How many neurons, layers?
  • How many Epochs?

10

slide-34
SLIDE 34

11

  • 3. Challenge: inferencing speed
slide-35
SLIDE 35

12

  • 4. Challenge: Safety & Acceptance
slide-36
SLIDE 36
  • safety first in ATC

12

  • 4. Challenge: Safety & Acceptance
slide-37
SLIDE 37
  • safety first in ATC
  • need to prove performance

12

  • 4. Challenge: Safety & Acceptance
slide-38
SLIDE 38
  • safety first in ATC
  • need to prove performance
  • regulator decides if system may be used operationally

12

  • 4. Challenge: Safety & Acceptance
slide-39
SLIDE 39
  • safety first in ATC
  • need to prove performance
  • regulator decides if system may be used operationally
  • we treat ANN as human, same tests as for ATController

12

  • 4. Challenge: Safety & Acceptance
slide-40
SLIDE 40

13

Example Images and Videos

slide-41
SLIDE 41
  • list a couple of sample sites and show actual video

14

Example Images and Videos

slide-42
SLIDE 42

15

Example Images and Videos

slide-43
SLIDE 43

16

slide-44
SLIDE 44

16

slide-45
SLIDE 45

17

slide-46
SLIDE 46

17

slide-47
SLIDE 47

18

slide-48
SLIDE 48

19

Future Work

  • Optimal Flight Level Prediction
  • Optimal Aircraft to Gate Assignment
  • AI Controller Assist
  • many potential new application in ATC
slide-49
SLIDE 49

Thank you! HEAD OFFICE
 19 Camelot Drive Ottawa, Ontario K2G 5W6
 PHONE 613 686 3988 TOLL FREE 1 866 799 1555 EMAIL info@searidgetech.com

Thank you for your time. I’ll be happy to answer any questions you may have.

slide-50
SLIDE 50

21

Challenge: Annotation

slide-51
SLIDE 51

Plattform Screenshots