Face detection on "Vertical video"

In my project I need a wide vertical angle face detection.
So I used widerange lens, rotate camera at 90, and get suficcient vertical FOV.
Horizontal FOV could be reduced to about 1/3 of vertical, and is not matter.

So I set windowing to ROI:
sensor.set_windowing((640, 300)

And want to rotate image and use haar cascade…

So question was:
Which is the right and memory efficient way to rotate image (not truncating dimensions, I need to get for example 300x640 from 640x300).
Or, maybe there if a way to “rotate” haar cascade (standard “frontalface”)?

Just use the rotation correction method:


I plan to add 90 degree rotation correction also to sensor in the future which will be less compute on the CPU. But, for now, we just have the above method.

Um, anyway, you might need to reduce the vertical res.