andred14 wrote: ↑
Sun Jul 05, 2020 10:20 pm
Edit Added top quote
The main goal for now is to just see where the bb or pellet is coming from. it may actually be a sniper bb or pellet gun from blocks away (I am in a high rise apartment and there are other high rises around)
ideally it would be great to get a thermal image so that I can see exactly where the projectile is coming from?
SuperZoom is very easy and possible with the HQ camera at a DIY cost
Old 35mm manual lenses with 2x or 3x adapters allow me to easily take pictures of the moon.
Why am I bringing this up?
You want as clear an image as possible to "catch" this person.
So I would set up more than one camera in the directions I think it may come from. Also they probably do it from a few set locations and over a few days you could possibly move the camera and zoom in directly on the corner or window it is coming from. Lastly it is possible with multiple cameras to potentially triangulate where they are within about a meter(way way above my skill level but I know It's possible)
I am Not a Coder(I hobble/hack code badly) so I use RPi_Cam_Web_Interface as it has everything built in.
I bring this up as people have used this software and custom coded some of the scripts so 1 camera triggers a series of cameras and it can easily do loop recording to grab frames from each camera before you technically hit record(forget the real name of this feature.)
Also no one has discussed the Pi's interframe option. This is only GPU possible so limited to 120fps at 640x480 or up to 1080p 30fps. This is a super low bandwidth option that only shows the vector differences(for all I know it is how different motion triggers work as it is from years ago)
Final thoughts. I would take 3 or 4 M12 lens Noir Camera knock offs with wide angle lenses and IR light from 1 to 5 feet. I would create a triangulation methodology with the cameras. Put RPiCamWeb on each Pi and link them. One gets motion they capture. Since I only need the vectors to triangulate I would explore the Vector Motion. Then Once that triangulation is done if I could I would use a computer controlled Pan Tilt to move the Superzoom HQ recording 1080P video at 30fps. Raw gets you 4k 12bit 10fps(11.3 if you follow what HermannSW has done). Just understand that RAW data is super processor intensive and requires Post Work. I am trying to get the pic of the person ASAP as they will gun & run. Also the triangulation method would allow you to narrow your Suoerzoom HQ for the next day or even the same night without a fancy computerized pan tilt. The vector/triangulation should be near realtime so if they stay put for 15 seconds you could manually pan zoom and grab the persons image.
I also forgot to mention StereoPi https://stereopi.com/
which can take stereo images from 2 cameras and could possible just give you triangulation inside a single Pi(2 Cameras is a Compute Module ONLY feature)
Way way above my level BUT research into triangulatuon using stereo Cameras on RasPi has been done here is an example Paper I found for Machine Vision
https://www.google.com/url?sa=t&source ... nn1MtKnmc9
https://www.google.com/url?sa=t&source ... EEETnFWEBg
Hope this helps.