Final Design

Our PPE A.I. Vending Machine integrates several subsystems into a cohesive solution:

Core Components

• Vision System: Camera and AI vision processing for PPE detection
• Control System: Central computing unit managing the entire workflow
• User Interface: Touchscreen display for user interaction
• Dispensing System: Vending machine for delivering PPE items
• Safety Gate: Access control mechanism based on compliance status

Data Flow

The system operates in the following sequence:

1. User approaches the machine and is detected by the camera
2. YOLO vision model analyzes the camera feed to detect PPE items
3. Results are processed by the computing system and displayed on the GUI
4. If compliant, the safety gate unlocks; if not, missing PPE is identified
5. User can request dispensing of missing PPE items through the interface
6. System dispenses requested items and re-evaluates compliance

The camera system serves as the "eyes" of our solution, providing visual data for AI processing:

Hardware Specifications

• Camera Model: Onsemi Ar0821 Camera Module
• Resolution: 8MP (4K)
• Frame Rate: Up to 60fps
• Dynamic Range: High Dynamic Range (HDR) capability
• Integration: Connected to Nvidia Jetson via MIPI CSI-2

Camera Placement

The camera is strategically positioned to capture users approaching the machine:

• Mounted at approximately 5.5 feet height
• Angled to capture full-body view of users
• Positioned to minimize lighting interference
• Field of view optimized for detection of all relevant PPE items

Image Processing Pipeline

• Real-time image acquisition at 30fps
• Resolution scaling to optimize processing speed
• Color space conversion for vision processing
• Pre-processing for noise reduction and image enhancement

Our solution employs YOLO (You Only Look Once) for real-time object detection of PPE items:

Model Architecture

• Base Model: YOLOv11n
• Neural Network: Convol
• Input Size: 640x640 pixels
• Inference Time: ~25ms on Nvidia Jetson

Detection Capabilities

The vision system can detect the following PPE items:

• Hard Hat
• Safety Glasses/Eye Protection
• High-Visibility Vest
• Protective Gloves
• Ear Protection (Plugs)

Performance Metrics

• Average Precision (AP): 85%
• Inference Speed: 30+ FPS on Nvidia Jetson

The GUI provides an intuitive interface for users to interact with the system:

Technology Stack

• Framework: PySide6 (Qt for Python)
• Backend: Python 3.8.10
• Styling: Qt Style Sheets (QSS)
• Resolution: 1024x768, responsive design (Depended on the physical screen size)

Interface Elements

• Live Camera Feed: Displays real-time video with PPE detection
• PPE Status Indicators: Color-coded buttons showing detection status
• Dispensing Controls: Touch buttons to request PPE items
• Safety Gate Control: Unlock indicator and override button
• System Status: Operational status and error messages

User Workflow

1. User approaches machine and views their image on the screen
2. PPE status indicators update in real-time
3. Compliant items show as green, missing items as red
4. User can press red buttons to request missing items/ or in automatic mode, the system will automatically request the missing items
5. System dispenses requested items
6. Once all items are green, safety gate unlocks automatically

Our solution integrates with commercial vending machine hardware for reliable dispensing:

Vending Hardware

• Base Unit: Modified commercial vending machine
• Control System: Two Avend SmartVend Kits
• Capacity: 24 slots across 6 vertical sections
• Item Types: Configured for diverse PPE items of varying sizes

Communication Protocol

• Primary Interface: HTTP REST API
• Endpoint Structure: Product-specific endpoints for dispensing

Dispensing Workflow

1. User requests item through GUI
2. Control system sends HTTP request to SmartVend unit
3. SmartVend controller sends corresponding signal to the Vending controller
4. Item is dispensed from the vending machine
5. Status is updated in the UI

The safety gate system provides physical access control based on PPE compliance:

Hardware Components

• Control Unit: ESP32 microcontroller
• Locking Mechanism: Electromagnetic lock (12V DC)
• Power Supply: Regulated 12V DC

Communication

• Protocol: Wired Serial Communication
• Commands: Lock/unlock
• Failure Mode: In case of power loss, the gate should be in the unlocked state

Safety Features

• Emergency override button for immediate unlock
• Automatic unlock in case of power or system failure
• Visual indicators for lock status
• Activity logging for all lock/unlock events
• Bluetooth broadcast for gate status

The actual safety gate will require specific designs for specific locations following proper protocols and regulations, so in out proof of concept we only created a simple gate controller indicator to demonstrate its function.

The entire system is powered by an Nvidia Jetson computing platform:

Hardware Specifications

• Model: Nvidia Jetson Orin Nano Developer Kit
• CPU: 6-core ARM Cortex-A78AE
• GPU: 1024-core NVIDIA Ampere architecture
• Memory: 8GB 128-bit LPDDR5
• Storage: 512GB SD Card

Software Environment

• Operating System: Ubuntu 20.04 LTS
• AI Framework: NVIDIA JetPack 5 SDK
• Vision Libraries: OpenCV, PyTorch, Ultralytics YOLO
• Application Stack: Python 3.8.10, PySide6

Performance Optimization

• CUDA acceleration for vision processing
• PyTorch, TorchVision combination optimized for utilizing the Jetson GPU
• Multithreaded architecture for parallel processing