We are building a self-scan checkout fraud detection system for supermarkets. The system must perform real-time person re-identification across multiple camera views (to track shoppers) and product detection/classification (to verify items, possibly offline). We plan to run all inference on a single Jetson module, processing 5 HD (1080p) camera streams.
Which jetson specification will be best suitable.
Hello @mohammad17,
Your project sounds very interesting.
I have a quick couple of questions:
-
Are all 5 camera streams going to be captured by the Jetson board or are they going to be streamed over the network to it ?
-
Do you already know which camera modules are you going to be using and what’s going to be their frame-rate capabilities ?
-
Any restrictions on form-factor, power consumption or operating temperature?
For instance, I would go with the Orin NX because of its price/capabilities ratio. You will be able to run AI smoothly on it and it has HW encoder and decoder in case you want to stream camera video to a user application. However, it has a limitation to the number of CSI cameras you can connect since you can only connect 4 cameras directly to the board, unless you use something else like GMSL or virtual channels, you will be limited on the number of cameras you can capture from.
If having 5 MIPI CSI cameras connected to the board is a hard requirement, then you might want to consider the AGX Orin board.
Would you be opposed to sharing more details about your use case and project?
We can even jump into a quick call (for free / no strings attached) to learn more about it and give you a better plan for building your MVP.
best regards,
Andrew
Embedded Software Engineer at ProventusNova
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.