i´m looking for the right hardware to make a blob detection/body tracking in an area of 10 to 10 meter. the cameras could be mounted in a heigth of around 3-5 meter. what is important to me is to keep track of the Person´s IDs, for instance when people start to move around, big people cover small people as kids etc. …
i guess the FUSION API with ZED360 is the right start. lets say i would go for the centralized workflow.
i would go for two ZED 2 cameras.
mount them at around 3m height, ( does the height affect the detection quality of the trained networks? because i guess the networks are trained from a certain angle right? would use COCO18 skeleton representation)
adjust each camera to angle downward slightly, aiming towards the center of the area or slightly overlapping in the middle.
as computer i could use a PC ( i7 6700 8-Thread CPU, 4.0 GHz | 32GB DDR4| Geforce RTX 3060 12GB DDR6 | 1TB SSD | Windows11 ) or an ZED Box NX Orin 16GB
i do not necessarily need all the body tracking data. mainly position with steady ID is important.
am I on the right track with this test setup?
could i maybe also have a camera setup of multiple monocular cameras as ZED X One, consequently have more perspectives?
If your goal is to only track people’s position, two cameras should be enough. I’d recommend 4 cameras for body tracking.
The angle of the camera can have a significant impact of the quality of the detection. As you can imagine, if the camera is top down, looking downward, almost all the body of the person is occluded.
But on the other hand, looking slight downward can help in certain configurations, in order to see a person occluded by someone else.
I think, placing the camera at 3m height can work very well.
You would need one ZED Box per camera, so I would recommend using a main computer with both cameras connected to it, if possible.
For the moment, only the body tracking module is available in the Fusion API, so you will need to use it. However, we plan to add the Object detection module in the next release of the ZED SDK (v4.2) which should work even better for you.
instead of going “de-centralized” with single ZED Boxes, you recommend one computer.
i read somewhere 4GB RAM per Camera should be taken to acccount, is that realistic? ( would go for ZED X Cams due to synchronisation instead of ZED i2 )
i wonder about connecting the cameras to the computer via GMSL2.
would it work to use 4 caermas with 4 usb-c ports?
nice about integrating object detection into the fusion module.
is there already a rough release date?
Sorry, I’won’t tell a release date, it’s not fixed yet, and my record is notoriously bad with estimations
*instead of going “de-centralized” with single ZED Boxes, you recommend one computer.
i read somewhere 4GB RAM per Camera should be taken to acccount, is that realistic? *
It’s VRAM and not RAM (though RAM is important too). That is realistic, keep in mind you also have your computer’s app to run.
would go for ZED X Cams due to synchronisation instead of ZED i2
i wonder about connecting the cameras to the computer via GMSL2.
would it work to use 4 caermas with 4 usb-c ports?
I said it just above but ZED X cameras are not compatible with PCs at the moment, the GMSL2 capture card is a limitation. Using 4 cameras on 1 computer might lead to USB bandwidth issues, be careful which ports you use (ideally back ports that are not adjacent, you may need a pcie usb extension card).
