Realtime Motion Detection Latency

[baesek]

Hi,
sorry, this is more of a question than an issue. But i did not find a Discussions area here and I have the feeling posting my question here could help finding an answer the best.

I am actually using a raspberry pi camera v1 as a realtime motion detection by using the PiMotionAnalysis Class and the start_recording function. I just output the number of movement vectors that are longer than zero to get a value for how much change is happening in an image.
It works quite well, but there is latency (around 400ms, it feels like) and I think the bottleneck is the video input.
I tried around for example using 640x480 at 90fps which improved latency but decreases a lot of detail in the detection as there are much less possible macroblocks / motion_vectors for the field of view.

Without thinking enough I now bought a Camera v2 module as I read that it offers 1280x720 (90fps) then I saw that this is 16:9 and therefore I would lose quite some of the image on the edges of the fisheye kind of objective I use.

Browsing the Internet I found out that there are many people somehow using the camera module at higher framerates. I wondered as I always looked at the docs https://picamera.readthedocs.io/en/release-1.13/fov.html?highlight=sensor%20Modes#sensor-modes to see whats possible.

Can you somehow help me to improve the setup?
I am wondering if there is any possible modification to the picamera lib to offer more fps at 1296x972 (4:3) for example.
Or if I can somehow output the motion_vector count much faster using a different approach than a python script which sends osc messages to a pure data patch.

Do you see my problem, do you know a good solution?

Thank you for any help.

best,
Paul