Build production-ready vision AI with Qt, QIM SDK, and Dragonwing IQ-9075

Figure 1. The FactoryPulse Vision AI pipeline in action -  the left display shows the safety-area alert with an intrusion-detected banner; the right display shows the defect-detection camera view with a live bounding box and confidence score drawn over the conveyor feed. 

Vision AI Use Cases in FactoryPulse

To demonstrate Vision AI on the Dragonwing IQ-9075 EVK, three use cases were implemented: 

• Defect Detection - a live video feed from a physical conveyor belt is analyzed by a Vision AI model to detect faulty objects; results are presented as a video overlay with bounding boxes. 

• Factory Digital Twin - defect detection events are surfaced with a red blinking alarm at the relevant station, together with a graph of faulty objects detected over time. 

• Worker Safety - the system detects people and their Personal Protective Equipment (PPE - safety helmet and vest) in restricted areas of the factory; results are displayed for the selected production line. 

Figure 3. Close-up of the Defect Detection area in the digital twin (light mode) — the machine-efficiency card in the top-right shows live performance, and the safety-area overlay is clearly visible in red. 

Figure 4. Operator dashboard in dark mode — the Vision AI tiles (top-right and bottom-right of the grid) show the Camera Live View and the running Defect Detection statistics, driven directly by the GStreamer pipeline. 

The demonstration includes 3 Vision AI models: 

• Edge Impulse's YOLO Pro model for defect detection 

Worker safety is supported through two TensorFlow Lite (TFLite) models: 

• A model for detecting protective gear (safety helmet and vest) 

• A model for tracking people near restricted areas 

All models are optimized for the Hexagon Tensor Processor (HTP) accelerators and run across the two HTP cores of the Dragonwing IQ-9075 EVK. 

Simplifying Edge AI: Declarative Video Pipelines with Qt Framework, Qualcomm IM SDK, and Edge Impulse

To deploy real-time video inference on the Dragonwing IQ-9075 EVK, we combined the Qualcomm IM SDK, Edge Impulse trained models, and the Qt Framework (QML). Our objective was to simplify Vision AI pipeline creation through a declarative approach, like how Qt streamlines user-interface development.

Build your first Vision AI pipeline

Combine Qualcomm IM SDK, Edge Impulse, and Qt for real-time video inference on the Dragonwing IQ-9075 EVK.

▶ [Start building with Qt + Qualcomm today: Request an evaluation license]

What is the Qualcomm IM SDK?

The Qualcomm IM SDK is a unified solution designed to accelerate multimedia and AI applications on Snapdragon platforms. Built on the open-source GStreamer framework, Qualcomm IM SDK provides a comprehensive catalog of plug-ins optimized for Qualcomm® hardware, including: 

• Qualcomm® Adreno™ GPU
• Qualcomm Spectra™ ISP
• Qualcomm® Hexagon™ NPU

These hardware-accelerated plug-ins enable: 

• Zero-copy buffer sharing 
• Efficient high-resolution video streaming 
• Direct data flow between camera, AI, and display components 
• Reduced CPU and memory overhead 

By abstracting low-level hardware details into standard GStreamer elements, Qualcomm IM SDK allows developers to build complex AI workflows that efficiently leverage edge hardware acceleration. 

Why Qt Framework is a Natural Fit

Qt Framework uses GStreamer as the default multimedia back end on embedded platforms. Since both Qualcomm IM SDK and Qt Framework rely on GStreamer, integrating them was straightforward. 

Qt Modeling Language (QML) provides: 

• Declarative UI design 

• Built-in animation and multimedia integration 

• Simplified application structure 

For complex pipelines, manually creating, initializing, and managing GStreamer elements typically requires significant C++ code. The objective in this project was to apply QML's declarative philosophy to AI pipelines, enabling developers to define AI workflows alongside UI definitions in a simpler and more intuitive way. 

This approach also adds platform abstraction, making it easier to port applications across platforms where different AI runtimes may be used. 

Learn more about Qt and QML. 

Defect Detection Pipeline with Edge Impulse

For the defect detection use case, we took the following steps:

• Trained a custom model: We used Edge Impulse to train a custom YOLO Pro model.
• Deployed the model: The Edge Impulse model was then deployed as a GStreamer element.

Because Edge Impulse models can be deployed as GStreamer elements, the Vision AI pipeline built with Qualcomm IM SDK could reuse the integration work directly. This reduced development friction and allowed the same GStreamer-based workflow to handle both hardware-optimized inference and model execution. 

Worker Safety Models

The worker safety use case includes two TFLite models: 

• Personal Protective Equipment (PPE) detection - helmet and vest 

• Restricted-area monitoring - person detection 

Both models are optimized for Hexagon HTP and integrated through Qualcomm GStreamer elements, enabling real-time inference directly on the Dragonwing IQ-9075 EVK. 

Optimized for Hexagon HTP

All Vision AI models in this solution were optimized for the Hexagon Tensor Processor (HTP) accelerators. They run across the two HTP cores of the Dragonwing IQ-9075 EVK, demonstrating how AI video pipelines can scale efficiently across the available acceleration resources. 

This architecture allows: 

• Concurrent AI workloads 

• Reduced CPU utilization 

• Efficient edge-inference performance 

From Demonstration to Reusable API: the Qt AI Inference API

Although FactoryPulse began as a demonstration, it evolved into the ideal environment in which to formulate a new Qt module and API for Edge AI development. The new module is built on top of Qt Multimedia - a cross-platform module for media playback, audio/video recording, and device management. 

At the time of writing, a proof-of-concept of the Qt module and API is available for developers to try. 

Try the Qt AI Inference API proof-of-concept. 

This represents a step toward improving the Qt Framework developer experience for building Edge AI applications on Qualcomm® IoT platforms. 

What Developers Can Take Away

This Vision AI implementation demonstrates: 

• How to combine Qualcomm IM SDK and Edge Impulse within a GStreamer-based pipeline 

• How Qt Framework can simplify AI pipeline orchestration 

• How to optimize models for Hexagon HTP 

• How to build efficient, hardware-accelerated Vision AI workflows on Dragonwing IQ-9075 EVK 

Ready to Get Started?

Simplify AI deployment with Qt APIs - Use declarative pipelines to build and scale Edge AI applications faster. Try the Qt AI Inference API proof-of-concept with the Dragonwing IQ-9075 EVK

Start building today: Request an extended evaluation license to build your own solutions with Qt and Qualcomm Technologies, Inc.

Train and deploy: Explore Edge Impulse workflows for Qualcomm to train, deploy, and run defect detection using Edge Impulse on Qualcomm hardware.

Watch it in action: Check out our webinar, FactoryPulse: Deploying Vision AI and Generative AI at the Industrial Edge with Qt, Qualcomm, and Edge Impulse.

Get hands-on: Follow our step-by-step tutorial, Build Along: Building Smarter AI UIs with Qt on Qualcomm Dragonwing.

Explore the tech: Learn more about Qt and QML to dive deeper into the technologies powering this solution.

Coming soon: Stay tuned for our upcoming technical deep-dive posts on GenAI + RAG!