Powered by Qualcomm Hexagon AI Engine
License: MIT
This reference design demonstrates a multi-camera, multi-model AI pipeline using the Tria VisionAI-Kit 6490. Built with Python, GTK, and GStreamer, this GUI-based application allows users to:
- 🔄 Run up to two AI models concurrently on separate camera inputs
- 🎥 Stream live camera feeds with overlays from selected AI models
- 📊 Visualize real-time system performance and thermal metrics
- 🚀 Leverage Qualcomm’s Hexagon DSP AI Engine for efficient, low-power inference
Currently supported ML pipelines include:
- Pose Detection
- Depth Segmentation
- Object Detection
- Image Classification
With potential for more!
The design integrates two utilities for performance monitoring:
- QProf (Qualcomm's common Linux profiler)
- psutil (a cross-platform system utility)
- CPU / Memory / GPU utilization
- LPDDR5, CPU, and GPU temperature graphs
- Track last N seconds (N=30 by default)
Here’s the demo setup we showcased at Embedded World 2025:
The VisionAI-Kit 6490 is a platform designed for edge-AI (multi) camera applications. Key features include:
- Qualcomm QCS6490 SoC
- Up to 4 MIPI camera inputs
- Integrated Hexagon AI Engine
- 8 GB LPDDR5 memory
⚠️ Note: although unlikely, HW could still be subject to change
This project is actively maintained. Contributions, feedback, and issue reports are welcome!
This project is licensed under the MIT License.
