Connecting MIPI CSI Camera Sensor to Hailo Pipeline

General Discussion
1

I’m working on a Hailo-based edge AI setup and I need some guidance on integrating a MIPI CSI camera sensor with the Hailo pipeline.

Specifically, I want to understand the correct workflow to:

  • Connect a MIPI CSI camera sensor (IMX 334) to a Hailo platform

  • Capture and process raw sensor data

  • Convert the raw Bayer output into a usable image format (RGB/YUV)

  • Feed the processed frames into a Hailo model (HEF file) for inference

I’m also trying to understand where the image processing should happen in the pipeline:

Sensor → ISP → GStreamer / V4L2 → preprocessing → Hailo inference

If anyone has experience with:

  • MIPI CSI sensor bring-up

  • ISP configuration (3A tuning, 3aconfig.json)

it would be really helpful if you could share a reference pipeline or architecture.

2

Hi @user1045,

Unfortunately Hailo-8 does not support MIPI.

Thanks,

3

What about in Hailo 10 or Hailo 15

4

Hi @user1045,

Only 15.

Thanks,

5

Hi @user1045 ,

The Hailo accelerators (Hailo-8/8L, Hailo-10H) are connected to the host over PCIe (USB3 is also possible for Hailo-10H).

Therefore the pipeline should look like:

  1. The sensor is connected to the host.
  2. The video pipeline in the host (possibly based on GStreamer) is responsible to read the sensor data and send it to the Hailo accelerator for inference.
  3. The Hailo device returns the inference results to the host.

The accelerators don’t support direct connection to a sensor.

On the other hand, Hailo-15H/L are SoCs so they are directly connected to the sensor. The application that runs on Hailo-15 handles both the vision and analytics. There is no need for an additional host.

6

I’m currently using an IMX334 sensor with Hailo15H.

I was able to find the sensor driver and XML file, but the required ISP tuning/configuration files such as 3aconfig.json and related profile files were missing for IMX334.

To proceed, I tried replicating the setup by referring to the existing IMX678 configuration and adapting it using the IMX334 XML values. Initially, there were pipeline/configuration errors, but I managed to resolve those issues.

At this point, the demo application under:

apps/case_studies/detection/detection_case_study

is running successfully, but the camera output is completely dark/black.

Because of this, I suspect there is still an issue with:

  • ISP tuning

  • exposure configuration

  • sensor gain/exposure limits

  • 3A configuration files

  • or HDR-related settings

Currently I’m stuck at the IMX334 configuration stage and not sure what the correct next step should be.

My questions are:

  1. Is there any official/reference 3aconfig.json or ISP tuning package available for IMX334 on Hailo15H?

  2. Does IMX334 fully support HDR on Hailo15H?

  3. Is there a recommended method to generate/adapt the ISP tuning files for a new sensor?

Any guidance or reference configuration for IMX334 would be really helpful.

7

Hi @user1045 ,

The Hailo Community currently supports our AI Accelerator products only.
For AI Vision Processors (Hailo-15) questions, please fill out the product inquiry form, and our dedicated team will get back to you shortly. Hailo AI Product Inquiry: Connect with Our AI Experts

Thanks,