Real-time stitching of videos from two IP connected camera

asked 2017-10-04 02:55:29 -0600

Umair Iqbal gravatar image

Hello guys, i am trying to stitch video feed of two camera connected through IP in an indoor office arrangement. I am using the opencv stitcher class for this purpose with optimization of different features, however video is not stitched instead output shows only the video from camera on left and ignores the video of camera on right. In my point of view this is mainly because

  1. in this indoor situation which involves tables, chairs, cabins and etc made of same material and identical in look and hence, features are matched wrong by feature matching algorithm.

  2. Furthermore, although there is lot of overlapping region between both images but most of it is plain tile floor and hence there are very few features available to match.

I have couple of more question related to this task in case if above problems are solved.

  1. There is a lag between both cameras (about 3 to 4 seconds) which i think is because of network router. How can that lag be compensated. Or in simple words, how can i get synchronous video feeds from the cams to process.

  2. I want to test my written algo against some standard dataset of video stitching to check if it is working fine in other situations or not. However i am unable to find any such dataset which have videos of same scene captured by two cameras (static).

Any help in this regard will be appreciated.

edit retag flag offensive close merge delete


As far as I know, there is not such data set yet, you can read the paper "deep image homography estimation". Maybe this paper could help you figure out better homography matrix?

tham gravatar imagetham ( 2017-10-04 03:13:24 -0600 )edit

Thank you Tham, i will read through the article if it is of any help to me.

Umair Iqbal gravatar imageUmair Iqbal ( 2017-10-04 03:22:19 -0600 )edit