♻️ RECLAIM

Recycling Classification and Live Automated Integrated Monitoring System

An end-to-end intelligent waste management ecosystem integrating Computer Vision, IoT edge hardware, and a centralized SaaS monitoring platform.

---

🎓 Final Year Project (FYP)
🏛️ Institution: New Era Institute of Vocational & Continuing Education (新纪元技职与推广教育学院)
👨‍💻 Author: CHEN JUN WEI (2470223-SEAIR)
👨‍🏫 Supervisor: Mr. LOO WEI HOU

---

Figure 1: RECLAIM Smart Bin full physical prototype and electronic control assembly.

🏆 Awards & Honors

RECLAIM has been showcased in national and international innovation competitions, achieving prestigious recognition:

• 🥇 GOLD AWARD — WOU INNOVATEX 2026 (Wawasan Open University — Innovation Excellence)
• 🥇 GOLD AWARD — BIICC 2026 (Borneo International Innovation Creativity Competition 2026, hosted by Universiti Malaysia Sabah - UMS)
• 🥈 Certificate of Appreciation — MIIX 2026 (Malaysia Invention & Innovation Expo 2026)

---

🌍 Alignment with UN Sustainable Development Goals (SDGs)

RECLAIM directly contributes to four key United Nations Sustainable Development Goals:

• 🟢 SDG 3 (Good Health and Well-being): Eliminates unhygienic waste overflow, suppressing pest breeding and bacterial spread in urban spaces.
• 🟠 SDG 9 (Industry, Innovation, and Infrastructure): Integrates edge AI and smart IoT sensing into modern municipal infrastructure.
• 🟡 SDG 11 (Sustainable Cities and Communities): Enables dynamic, data-driven municipal collection routes that reduce greenhouse gas emissions and fuel waste.
• 🟤 SDG 12 (Responsible Consumption and Production): Automates source separation (Aluminum, Glass, Paper, Plastic) and promotes circular economy habits.

---

📈 Background & Problem Statement

Why RECLAIM?

Traditional municipal waste systems suffer from severe structural inefficiencies:

1. Low Recycling Rate: Malaysia's national recycling rate sits at 31%, lagging behind regional peers like South Korea (49%), Singapore (59%), and Taiwan (60%).
2. Escalating Municipal Spending: Malaysia's solid waste and environmental cleaning expenditure has escalated dramatically from RM 3.0 Billion in 2020 to an estimated RM 7.53 Billion by 2025.
3. Static Collection Routes ("Empty Trips"): Rubbish collectors follow fixed timetables regardless of bin capacity, burning unnecessary fuel for empty bins while high-density bins overflow.

Figure 2: Benchmark comparison of recycling rates and Malaysia's 2020–2025 waste expenditure trend.

---

💡 The RECLAIM Solution & Architecture

RECLAIM delivers an end-to-end closed loop from waste detection to automated sorting, real-time telemetry, and citizen education:

Figure 3: Closed-loop ecosystem: Detection ➔ Sorting ➔ AI Coaching ➔ Long-term Habit Formation.

System Data Flow

flowchart TD
    subgraph Vision ["🖥️ Vision Processing (Laptop / PC)"]
        Cam[Webcam Feed] --> YOLO[YOLO11 Detection Model]
        YOLO --> Filter[7/10 Frame Stabilization Logic]
    end

    subgraph Edge ["🍓 Edge Gateway (Raspberry Pi 3)"]
        FlaskAPI[Flask Communication Server :5000]
        GPS[ATGM336H GPS Module]
    end

    subgraph Actuation ["⚡ Control & Physical Execution (Arduino Uno)"]
        Uno[Arduino Uno Controller]
        ToF[VL53L0X Laser ToF Sensor]
        Servos[MG995 & SG90 Servo Motors]
        Lid[Rotary Platform & Drop Lid]
    end

    subgraph Server ["🌐 Centralized Platform (Django + MySQL)"]
        Django[Django Web Server :8000]
        DB[(MySQL Database)]
        Groq[Groq API - Llama 3.3 LLM]
        SMTP[Gmail SMTP Service]
    end

    %% Connections
    Filter -->|HTTP GET /action| FlaskAPI
    FlaskAPI -->|USB Serial @ 9600 Baud| Uno
    Uno -->|PWM Signals| Servos
    Servos --> Lid
    Uno -->|I2C Read Distance| ToF
    Uno -->|Capacity Level String| FlaskAPI
    GPS -->|UART NMEA Data| FlaskAPI
    FlaskAPI -->|HTTP POST /sbin_api/report-trash/| Django
    Django <--> DB
    Django <--> Groq
    Django -->|Send Alert Email| SMTP

Loading

---

✨ Key System Features

• 👁️ YOLO11 Multi-Class Waste Detection: Classifies four primary recyclable categories (Paper, Aluminum, Plastic, Glass) with high accuracy.
• 🛡️ 7/10 Frame Stabilization Logic: Implements a rolling 10-frame buffer requiring $\ge 7$ unanimous detections before triggering mechanical motors, eliminating camera noise and glare errors.
• 🎯 VL53L0X Time-of-Flight Laser Sensor: Laser-based millimetric distance measurement inside enclosed bins, resolving the sound-wave bounce inaccuracies of traditional HC-SR04 ultrasonic sensors.
• 📍 Real-Time GPS Asset Tracking (ATGM336H): Live geolocation telemetry to prevent bin theft and optimize fleet routing.
• 🌐 SaaS Fleet Management (Dual-Ownership Model):
  • Private Bins (Customer-owned): Assigned to specific corporate/campus administrators (e.g. Admin Office, Cafeteria).
  • Public Bins (Government-owned): Monitored for municipal locations (e.g. MRT stations, public parks).
• 🚨 Dual-Channel Notification Engine: Immediate on-screen warning badges combined with automated SMTP email alerts dispatched when any compartment reaches $100%$ capacity.
• 🤖 SBIN Assistant (Groq Cloud LLM): Integrated conversational chatbot powered by Llama-3.3-70b-versatile that answers user queries on recyclability and provides live bin capacity status without loading the edge GPU.
• 📹 Remote Camera & Detection Control: Web dashboard toggle allowing administrators to remotely turn the camera detection script on/off.

---

📊 Model Training & Hardware Testing Results

The custom dataset was annotated via Roboflow and augmented across diverse lighting and perspective variations. YOLO11 was trained over 130 epochs.

Training & Validation Loss Convergence

Validation Confusion Matrix

Metrics Summary:

• mAP@50: 0.98
• mAP@50-95: 0.82
• Precision & Recall: Both stabilized above 0.95
• Real-world Classification Accuracy: 85% – 90%
• Mechanical Sorting Accuracy: 100% across 80 physical test trials (20 per category: Paper, Aluminum, Plastic, Glass).

---

🔮 Future Outlook & Roadmap

1. Gamified Rewards System: Citizen incentive app where proper waste sorting earns points redeemable for transit vouchers or retail discounts.
2. Cleaner Staff Mobile App: Mobile dispatch application with automated route optimization algorithms based on live fill percentages.
3. Solar Energy Harvesting: Integrated solar panels and battery storage for 100% off-grid operation in outdoor public parks and bus stops.

---

🚀 Quick Start & Installation

1. Prerequisites

• Python 3.10+ installed
• MySQL Server (5.7 or 8.0+) installed and running
• Arduino IDE (for flashing reclaim_arduino.ino)
• Web camera or laptop camera

2. Clone the Repository

git clone https://github.com/chenjunwei1230-dotcom/RECALIM.git
cd RECALIM

3. Setup Virtual Environment & Dependencies

# Create and activate virtual environment
python -m venv venv
# On Windows:
.\venv\Scripts\activate
# On Linux/macOS:
source venv/bin/activate

# Install required Python packages
pip install -r requirements.txt

4. Configure Environment Variables

Copy .env.example to create your local .env:

copy .env.example .env

Open .env and fill in your MySQL credentials, Gmail SMTP settings, and Groq API Key.

5. Database Migration & Superuser

python manage.py makemigrations
python manage.py migrate
python manage.py createsuperuser

6. Run the Services

1. Start the Django Web Platform:

python manage.py runserver 0.0.0.0:8000

2. Start the YOLO Vision Detection Module:

python scripts/yolo_detector.py

3. Deploy Raspberry Pi Gateway (on Pi 3):

python hardware/raspberry_pi/reclaim_raspberry_pi.py

4. Upload Arduino Code: Open hardware/arduino/reclaim_arduino.ino in Arduino IDE, install Adafruit_VL53L0X library, and upload to the Arduino Uno.

---

👥 Contributors & Academic Credits

• Developer: CHEN JUN WEI
• Academic Supervisor: Mr. LOO WEI HOU
• Faculty: Department of Software Engineering & AI, New Era Institute of Vocational & Continuing Education

---

_{Developed for Final Year Project (FYP) • Academic Year 2025/2026}