Dear everyone,
I have to develop a software which detect when a video is wrongly rotated (for example this one https://youtu.be/F4zqCHfzudw?t=25s)
We cannot assume things like there are/aren't people or there is/is not sky... I have been thinking about how to address this issue:
- Analyse the global movement in the video using differences between frames and HOGH transform... But this one doesn't work very good on a static videos.
- Try to analyse which kind of image is: people, landscape, city and then look for characteristic features of each situation...
- Look for text ...
I am quite stuck in this task... Could give me your opinion to try to find the way to address this ?
Thank you so much!