Ask Your Question
0

Real-Time video stitch (moving cameras)

asked 2016-12-21 15:00:00 -0600

thealse gravatar image

updated 2016-12-22 02:20:35 -0600

Hi, I have just started to use openCV and I am very new to image/video processing and stitching so please bear with me while I explain what I am trying to accomplish.

  1. I plan to purchase 3 HERO5 Black and connect them to a PC via HDMI
  2. For this I plan to use the following Capture Card, http://www.magewell.com/pro-capture-q...
  3. I plan to buy the following NVIDIA Graphic Card Nvidia Titan X, for processing the images
  4. The cameras position/angle between each other will be fixed.The cameras will be moving with respect to the world, so the scene will be changing.

I would like to use openCV and preferably python to stitch real-time videos from these cameras into one video.

So first of all, I must start by asking

  • Is the hardware I plan to purchase sufficient to solve the task? (is GoPro a good choice to do live video editing?)
  • Does OpenCV support reading video feed from this capture card/HDMI input?
  • Should I focus on using Linux or Windows, what platform is better supported by OpenCV? (hopefully not MAC)
  • Where/How should I start with the programming? Are there simple OpenCV examples for reading HDMI feed and stitching out there as I believe someone else has already tried to solve this problem? I found the following link, but the project is too complicated for me to start with directly, https://github.com/lukeyeager/StitcHD (this seems to difficult to start with)

Any help is appreciated.

edit retag flag offensive close merge delete

Comments

Where/How should I start with the programming?

since chances are low, to use opencv's VideoCapture for this, you should try their SDK, which unfortunately won't be in python ;(

once, you can acquire images, we can help you further.

berak gravatar imageberak ( 2016-12-22 02:43:44 -0600 )edit

If you are planning to make panorama from moving camera or video, then you must have a look at this tutorials here

vin gravatar imagevin ( 2017-06-05 10:12:49 -0600 )edit

A link is not an answer. Repost as comment

LBerger gravatar imageLBerger ( 2017-06-05 11:29:51 -0600 )edit

1 answer

Sort by ยป oldest newest most voted
1

answered 2016-12-21 18:01:14 -0600

Tetragramm gravatar image

Is the hardware I plan to purchase sufficient to solve the task?

Not enough information.

Does OpenCV support reading video feed from this capture card/HDMI input?

Almost certainly not, but if you can get a pointer, you can construct an OpenCV Mat around it and everything is fine.

Should I focus on using Linux or Windows, what platform is better supported by OpenCV? (hopefully not MAC)

Either or. Both are well-supported. Your comfort with it makes much more difference.

Where/How should I start with the programming?

Start with the Python Tutorials. Get comfortable with OpenCV, learn how it works ect. Then the Image Stitching tutorial. It's in C++, so you'll need the previous tutorials to understand how to translate it to Python. The rest of what you need is in the documentation on Stitching. The basic tutorial is not for video, but by the time you've worked through them all, you'll know how to read the documentation and have a better idea of what to do to make it fast for video.

Good Luck.

edit flag offensive delete link more

Comments

@Tetragramm, Thanks for the reply, some question though, I mentioned which camera I plan to use, is that not enough information? Also using a capture card, couldn't I do

cap = cv2.VideoCapture(0)
cap = cv2.VideoCapture(1)
cap = cv2.VideoCapture(2)

To read from all the cameras? (please also see updated question with more info)

thealse gravatar imagethealse ( 2016-12-22 00:22:58 -0600 )edit

Stitiching can be done when camera is rotated arournd its optical center

LBerger gravatar imageLBerger ( 2016-12-22 10:06:17 -0600 )edit

The information I'd need is how good a programmer you are, and how good you are at optimizing. A bad programmer can make any hardware seem slow.

And no, unless they mask the cameras to appear as standard plug-n-play cameras, then that's not going to work. As berak said, you'll need to use their SDK.

Tetragramm gravatar imageTetragramm ( 2016-12-22 11:27:03 -0600 )edit

Question Tools

1 follower

Stats

Asked: 2016-12-21 15:00:00 -0600

Seen: 4,478 times

Last updated: Jun 05 '17