Hi everyone,
I’m currently working on an academic project using a Raspberry Pi 5 and a Hailo-8 AI accelerator. I’ve been experimenting with the hailo-appsinfrastructure, but I’ve hit a design crossroads regarding visualization and control.
I want to create a custom overlay. Specifically:
-
Adding Custom Graphics
-
Renaming detected classes (e.g., changing “person” to “id number”).
-
Changing bounding box colors based on specific logic
-
Extracting the $(x, y)$ coordinates of the target center
The Current Dilemma
I’ve successfully run the pre-built GStreamer pipelines, but hailooverlay is a pre-compiled C++ element. While it’s incredibly fast, it feels “rigid” for a custom HUD.
I have tried looking into my options and want to know if someone have done any of them ?
1.Modify the existing helper pipelines to replace hailooverlay with an identity element using signal-handoffs=true. I would then use a Python callback with OpenCV to draw on the buffer.
2.Bypass GStreamer strings entirely and use a higher-level SDK (like DeGirum or a direct HailoRT Python loop).
Looking forward to hearing how you guys handle custom UI and hardware triggers in your projects!
#RaspberryPi5 hailo8 gstreamer #OpenCV #ComputerVision #TargetingSystem