Vision Acceleration on RPi/Hailo 8L?

General
,
1

Hi,

I’m developing a project that is based on the Raspberry Pi 5 and Hailo 8L, it combines frames from two wide angle cameras and from that needs to create a virtual pan/tilt/zoom cameras. Think of a CCTV type scenario that can follow people around and you’ll be close.

I’m relatively new to computer vision tasks and most of my experience has been with the Luxonis camera system which includes a Movidius chip to do a lot of vision processing in hardware rather than making the Raspberry Pi take the hit.

What I’m trying to figure out is if I can use the Hailo 8L in a similar way and perform CV functions like perspective warping, and similar, on the 8L rather than the Pi’s CPU? I’m trying to maximise the resolution I can process and figured if there’s the capability I should really be using it!

I’ve a few decades experience as a software engineering behind me so I’m good with the actual coding side of things, I’m just completely overwhelmed with this level of computer vision and machine learning if I’m honest, so if anyone could give me pointers regarding computer vision tasks on the 8L I’d appreciate it! If not, if there’s another chip that would be better suited I’d be happy to hear about it too.

Thanks,
Keegan

2

Welcome to the Hailo Community!

The Hailo-8L is an accelerator designed to run the compute intensive part of neural network inference. To make use of it, you must describe your function as a network (e.g. Pytorch or Tensorflow) using layers supported by the Hailo Dataflow Compiler. Have a look at the Hailo Dataflow Compiler User guide for a complete description of the supported layers.

There is no low level API that would allow you to just program any function you want. The Hailo devices are not general purpose GPUs.

3

Thanks for the clarification, I expected this to be the case but wasn’t sure if there was something on board for pre- or post-processing that I could use.