Therefore, am now looking for project ideas related to it.
Would be very grateful if I could get some recommendations from you guys. I should mention that the project should showcase about 6 months of hardwork from a group of 3 people. Therefore, I require an idea which can be done in that timeframe.
Thanks in advance.
Plus if you check out this list you might see some other ideas that are emerging from the industry: https://twitter.com/i/lists/1589644342746021890 or https://twitter.com/i/lists/1588109045902319616
All the best!
The answer depends on what kind of CV you're looking for. As for the old-fashioned CV I think that visual odometry, bundle adjustment and dense reconstruction are great things to play with and they have practical applications.
One hard-ish problem I used to work at is city-scale 3D reconstruction from photos. You can have street-level image data you collected to get the high detail and some aerial pictures to get the consistency for wide-area. A problem of matching street-level and aerial pictures is even interesting from research point of view (keyword is wide baseline image matching). With this you can build something similar to what Google Map has where you point your phone on the street and you get overlay directions. And if you call it mixed-reality metaverse, you can start a startup 6 months ago.
As for the NN-based I'll leave that for others.
[1] https://taylor.town/livecount
(Skip to 0:59 if you want to see the actual demo without the configuration.)
In order to run at a high framerate, I implemented 100% of the code from scratch in Swift. There's no OpenCV or any third-party dependencies in there.
I'm not working on it anymore, so if y'all wanna use it is as a starting point for improvements, send me an email at hello@taylor.town