presentation: simulating high quality video from still images

13
HIGH QUALITY SIMULATED VIDEO FROM STILL IMAGES Advisor – Dr. Shay Har-Noy | Technical Lead – David Schmidt 1 presented by Alexander Chan & Nima Hashem ViaSat Proprietary Information

Upload: alexander-chan

Post on 25-Jul-2015

879 views

Category:

Engineering


3 download

TRANSCRIPT

Page 1: Presentation: Simulating High Quality Video from Still Images

HIGH QUALITY SIMULATED VIDEO FROM STILL IMAGES

Advisor – Dr. Shay Har-Noy | Technical Lead – David Schmidt

1

presented by Alexander Chan & Nima Hashemi

ViaSat Proprietary Information

Page 2: Presentation: Simulating High Quality Video from Still Images

|Typical UAV System

ViaSat Proprietary Information

1. UAV captures video of ground in real time

2. Data transmitted via airborne modem3. Data passes through 10 mb/s data link4. Video received through ground modem

and viewed

(2)

(3)

(4)

(1)

Page 3: Presentation: Simulating High Quality Video from Still Images

|Motivation: Pen or Marker?

ViaSat Proprietary Information

3

Problem:Detecting small features

in video is difficult.

Resolution: Use high resolution still

images to create simulated video.

Page 4: Presentation: Simulating High Quality Video from Still Images

| Comparing Video and Still Images

ViaSat Proprietary Information

4

Spatial versus Temporal Resolution

HD Video 1920x1080 = 2 Mp per frame. 3.2

pixel/inch Still Images

4743 x 3162 = 15 Mp per frame. 7.9 pixel/inch

Page 5: Presentation: Simulating High Quality Video from Still Images

| Approach

ViaSat Proprietary Information

5

(1)

(2) (4)

(3)

Page 6: Presentation: Simulating High Quality Video from Still Images

|System Overview

ViaSat Proprietary Information

6

Page 7: Presentation: Simulating High Quality Video from Still Images

|Image Stitching Algorithm

ViaSat Proprietary Information

7

Page 8: Presentation: Simulating High Quality Video from Still Images

|Image Stitching Visualized

ViaSat Proprietary Information

8

Extracted Feature Points (SURF) Paired Feature Points Remain (FLANN Matching & RANSAC)

Stitched Mosaic (Feature Mapping and Blurring)

A BA & B

Page 9: Presentation: Simulating High Quality Video from Still Images

|Demonstration

ViaSat Proprietary Information

9

Zooming into video for visual feature detection

Synchronization between video and image windows

Page 10: Presentation: Simulating High Quality Video from Still Images

|Current Limitations and Future Steps

ViaSat Proprietary Information

10

Increase robustness of stitching algorithm

Increase accuracy of time relation between video feed and image viewer

Optimize CPU consumption throughout system

Page 11: Presentation: Simulating High Quality Video from Still Images

|What We Learned

ViaSat Proprietary Information

11

Image Processing and User Interface

Project Management and Spiral Development

“Integration takes longer than development”

Integrating Open Source Libraries (OpenCV, IJG Library, and existing EnerView

system)

Page 12: Presentation: Simulating High Quality Video from Still Images

|Conclusion & Applications

ViaSat Proprietary Information

12

Page 13: Presentation: Simulating High Quality Video from Still Images

|Acknowledgements

ViaSat Proprietary Information

13

Dr. Shay Har-Noy, Dave Schmidt, Steve Gardner

Fran Abrams and the ViaSat HR team

OpenCV Community, IJG Community

Kevin, Andy, Brian, Ricky, and our intern family

The Antarctic Penguins