Computer Vision Conveyor System

The initial approach was simple: process one SIM card at a time. However, a pivotal discovery changed everything. The chosen machine learning model, Edge Impulse's MobileNetV2 SSD FPN-Lite, was capable of detecting up to 10 objects in a single frame.

"Why process one card per frame when the model can see ten?"

This insight sparked a complete redesign. The solution was to build a custom 9-card batch dispensing system, controlled by an ESP32. This hardware innovation, combined with the power of the ML model, increased the system's throughput by 900%.

The system is a multi-layered solution that seamlessly integrates hardware, edge intelligence, and cloud automation. Follow it from the dispenser to the dashboard.

The same task, two very different realities. Switch between the manual process and the automated system.

The transformation was dramatic. The new AI-powered system delivered massive improvements across the board.

With a development cost of around $10,000, the system achieved a payback period of just 3 to 4 months, with a projected five year savings of over $140,000.

Building this system required mastering 16+ technologies with no formal training. Here is how it progressed from basic Python scripting to production IoT development, in three phases.

Two patterns did the heavy lifting: real-time streaming to the browser, and confidence-based routing between edge and cloud.

// Socket.IO WebSocket Events
socket.on("frame_update", (data) => {
  // Live feed at 30 FPS
  setCameraFrame(data.frame);
});
 
socket.on("shutter_sound", () => {
  // Audio feedback
  playShutterSound();
});

# Smart confidence-based routing
if edge_confidence > 0.95:
    return edge_result  # 50ms
elif edge_confidence > 0.80:
    return await cloud_verify()  # 2s
else:
    return await cloud_process()  # 3s

The current system is a production-validated success. The next logical step is to evolve the hardware from a prototype to a polished product. The future of this project lies in custom PCB design.

The goal is to design and manufacture a custom Printed Circuit Board that integrates the core electronic components, the Raspberry Pi Compute Module, the ESP32, motor drivers, and power management circuits, onto a single, compact board. This will not only shrink the device's physical footprint but also dramatically increase its reliability and robustness, making it suitable for deployment in harsh industrial environments. This step marks the transition from a one-off solution to a scalable, manufacturable product.