This is my setup. I have a Raspberry Pi 5 and a PCIe card with 4x HAILO-8 chips. There can be up to 6 HAILO-8 chips on the card, like this one. I use a PineBoard to connect the PCIe card to the PCIe slot on Pi 5. The whole idea here is to run one neural network on one HAILO-8 chip. Since there are multiple HAILO chips, it can run inferencing on multiple pipelines simultaneously. For example, for real-time object detection, I want to run 2x YOLOv8 models on 2 chips, another 2 ResNet models on another 2 chips. So I need 4 cameras (or more if I have 6 HAILO chips).
But, what is the best camera strategy for this? It doesn't have to be just USB cameras. If I go for all 4x USB cameras, I certainly need USB hubs. Or, use instead 2x IP cameras and 2x USB 3? Would the PCIe x1 slot handle that much data? Any suggestions on this?
But, what is the best camera strategy for this? It doesn't have to be just USB cameras. If I go for all 4x USB cameras, I certainly need USB hubs. Or, use instead 2x IP cameras and 2x USB 3? Would the PCIe x1 slot handle that much data? Any suggestions on this?
Statistics: Posted by Matt_W_ — Sun Dec 29, 2024 4:37 am — Replies 2 — Views 44