VisionSort-RPi

An AI-Powered Conveyor Belt Sorting System Using Raspberry Pi + YOLOv8 + Hailo

A real-time bolt-and-nut sorting system powered by computer vision and embedded AI. This project blends hardware engineering with object detection and real-world automation — combines hardware engineering, object detection, and creative problem-solving to efficiently sort Bolts and Nuts.

---

Overview

VisionSort-RPi is an edge-deployed conveyor belt system that showcases real-world applications of embedded AI.
Built using Raspberry Pi 5, YOLOv8, and the Hailo AI accelerator, it automates object classification and sorting tasks with high accuracy and speed.

---

🖼️ 3D Model & Real Conveyor Sorting System

3D CAD Model

Real Conveyor System

Side-by-side view of the designed CAD model and the actual built system with camera and flapper mounted.

Key Features

• ✅ Real-Time Object Detection with YOLOv8 + Hailo for bolts and nuts
• ✅ Stepper + Servo Motor Integration for precise movement and sorting
• ✅ Edge Deployment on Raspberry Pi 5 with hardware-accelerated inference
• ✅ Custom Dataset with over 9,000 annotated training images
• ✅ Optimized for speed using neural network acceleration (Hailo RT)
• ✅ Fully Autonomous Sorting with no human-in-the-loop

---

Tech Stack

Category	Tools & Hardware
🧠 CV/AI	YOLOv8, Roboflow, Python, OpenCV
⚙️ Embedded HW	Raspberry Pi 5, Hailo AI Kit, NEMA 17, MG996R Servo
📷 Camera	Raspberry Pi HQ Camera + 16mm 10MP Telephoto
🔌 Communication	GPIO, PWM, I2C
📦 Dataset	Roboflow custom-labeled (9000+ images)

---

Mechanical Components & CAD

This project includes custom-designed mechanical components modeled in Creo Parametric and fabricated using 3D printing. Key components include:

• Flapper mechanism for nut/bolt sorting
• Servo mounting bracket
• RPi HQ camera holder

🔗 View CAD designs & prints in CAD/
📄 Download full BOM: BOM.xlsx
📤 External CAD repo: GrabCAD – VisionSort-RPi

📦 Download 3D printable CAD models: ➡️ GitHub Release: CAD v1.0

---

Project Architecture

Hardware Design

• Conveyor belt driven by NEMA 17 stepper motor
• Servo-controlled deflector arm for object redirection
• HQ Camera with a 16mm telephoto lens for accurate detection
• All powered and controlled via Raspberry Pi 5

Data + AI

• Pre-annotated dataset from Roboflow
• Bounding box labels with nut and bolt classes
• Trained on HPC using YOLOv8 custom config
• Inference optimized with Hailo RT SDK

---

Software Pipeline

1. Camera Frame Capture →
2. Image Preprocessing (OpenCV) →
3. YOLOv8 Inference (Hailo AI) →
4. Object Classification →
5. Motor Control Signal →
6. Stepper + Servo Movement

---

Setup Instructions

Hailo AI Environment Setup

This project utilizes Hailo’s AI acceleration platform for deploying YOLOv8 models on the Raspberry Pi 5.
You can find the environment setup in their official repo.

Installation Steps

git clone https://github.com/hailo-ai/hailo-rpi5-examples.git
cd hailo-rpi5-examples
./install.sh

Every time you open a new terminal session, activate the environment:

source setup_env.sh

---

Running the Conveyor Sorting System

Run the real-time detection and sorting application using:

python detection/detection.py \
  --labels-json resources/nut_bolt-labels.json \
  --hef-path model/nut_bolt_model.hef \
  --input rpi

Press Ctrl+C to stop the system gracefully.

---

Hardware Wiring Diagram

The wiring for the stepper and servo motor system was meticulously crafted for precise control and safety.
Key components include the TMC2208 stepper driver, MG996R servo motor, buck converter, and external 12V power supply.

Wiring Summary

Component	Raspberry Pi GPIO	Notes
TMC2208 - STEP	GPIO17 (Pin 11)	Step signal
TMC2208 - DIR	GPIO27 (Pin 13)	Direction signal
TMC2208 - EN	GND	Must be LOW to enable
TMC2208 - VIO	3.3V	Logic level
TMC2208 - GND	GND
MG996R - SIGNAL	GPIO12 (PWM)	Servo control via PWM
Servo Vcc	Buck Converter 5V	Ensure stable 5V using multimeter
Buck Converter Input	12V external supply	Powers motors + regulated 5V output

📎 Full schematic is detailed in docs/Wiring_Setup.pdf

---

Training the YOLOv8 Model

The model used in this project is a fine-tuned YOLOv8n trained on a custom Roboflow dataset (9000+ images).

Training script used: training.py

• Based on YOLOv8n.pt pretrained weights
• Trained to detect only two classes: bolt and nut
• Handled augmentation, batch balancing, and early stopping

---

Model Conversion for Hailo Deployment

YOLOv8 .pt model was converted to Hailo-compatible .hef format using the Hailo Dataflow Compiler (DFC).

Conversion steps are documented in Convert_YoloV8_to_HEF.ipynb

Pipeline

• .pt → .onnx via Ultralytics export
• .onnx → .hef using Hailo’s hef-generator script

Guide followed: Guide to Using Hailo DFC

---

System Demonstration

Real-time sorting demo using computer vision and Raspberry Pi:

🎥

---

What's Next?

• Add camera calibration + lighting normalization
• Expand dataset to include more object categories for diverse sorting applications.
• Integrate IoT for remote monitoring.

---

📩 Contact

I’m excited to connect and collaborate!

• Email: gbrohiith@gmail.com
• LinkedIn: https://www.linkedin.com/in/rohiithgb/
• GitHub: https://github.com/GBR-RL/

---

📚 License

This project is open-source and available under the MIT License.

---

🌟 If you like this project, please give it a star! 🌟