💡 WellOffice - Enhancing Room Experience with Smart Lighting

An IoT-Based Smart Lighting System for Enhancing Focus and Relaxation in Office Spaces

Authors: Janice Sakina Akiza Nizigama, Tew Chuan Le, Eugenio Turcott Estrada, Gustavo Lucio Sepúlveda,
University: Aalborg University Copenhagen
Supervisor: Yan Kyaw Tun
Project Period: Fall 2024 (E24 – IoT-based Systems and Architectures)
Repository: gusls02/P5_Prototype1# & forked ver. JSanizi/WellOffice

🎥 Demo Video

---

📖 Overview

WellOffice is a smart lighting system designed to enhance focus and relaxation in a work office by adjusting the lighting based on human activity and presence.
The project integrates machine learning, IoT devices, cloud storage, and real-time adaptive lighting control to deliver a personalized lighting environment.

According to research and system testing, lighting influences attention and comfort.
WellOffice uses this insight to automatically adjust:

• ∼4300 K for focus
• 2700 K for relaxation
• 3500 K as the default state
• 15% brightness when no one is in the room

The system determines the user’s activity through camera input and adapts the light accordingly.

---

🧠 Project Architecture

System Components

1. Human Activity Recognition (HAR)

   • Implemented using pretrained PyTorch HAR models.
   • Determines whether the user is in a relaxed or focused state.
   • Models stored in HAR_models/ and src/lightbulb/bestHAR.pt.

2. Camera & Cloud Integration

   • src/camera/cameraS3.py
     • Captures images from the Raspberry Pi camera.
     • Uploads captured images to AWS S3 for further detection processing.
   • S3 notifications are used to trigger the detection workflow.

3. Object Detection (Roboflow + YOLOv11)

   • The system sends images to Roboflow for:
     • Human detection
     • Light bulb detection
   • Roboflow returns bounding boxes and detection results.

4. Light Bulb Logic & Control
   Located in src/lightbulb/:

   • human_model.py
     • Runs HAR inference to classify the user’s activity.
   • lightbulbs_model.py
     • Handles logic for interpreting detected light bulbs.
   • eulcidian_process.py
     • Calculates Euclidean distances between the detected human and each light bulb.
     • Determines the bulb closest to the user.
   • microcontroller_logic.py
     • Applies lighting rules based on activity and distance.
     • Sends commands to Philips Hue bulbs through a Zigbee gateway (e.g. ConBee II).
   • config.py
     • Configuration for paths, topics, device IDs etc.
   • lightbulbslogic.json
     • Stores light bulb positions, IDs, and behavior mappings.
   • Default lighting behavior from the report:
     • Default mode: 3500 K, 85% brightness
     • Relax mode: 2700 K, 85% brightness
     • Focus mode: 4300 K, 100% brightness
     • No-human mode: 3500 K, 15% brightness

5. Configuration & Utilities

   • src/config/datasetCleaner.py – cleans and preprocesses datasets.
   • src/config/datasetIndexer.py – indexes and structures datasets.
   • src/config/testMQTT.py – verifies MQTT connectivity.

6. Test & Evaluation Images

   • test_pictures/
     • Contains labelled images (e.g. janice1computer.jpeg, gustavoworking.jpeg, lightbulb1on.jpeg, lightsoff.jpeg)
     • Used to validate detection performance and end-to-end logic.

---

Example Flow – From Detection to Light Control

1. cameraS3.py captures an image and uploads it to S3 (or reads from test_pictures/).
2. Roboflow performs object detection on the image:
   • Detects humans
   • Detects light bulbs
3. Detection results are sent back to the system.
4. eulcidian_process.py computes which bulb is closest to the detected human.
5. human_model.py classifies the user’s activity (focused, relaxed, idle).
6. microcontroller_logic.py decides:
   • Which bulb(s) should turn on/off.
   • What brightness and CCT to apply (e.g. ~4300 K for focused work, warmer for relaxation).
7. Commands are sent to the smart lighting system (e.g. Philips Hue via Zigbee gateway).

---

⚙️ Installation

Requirements

• Python 3.10 or newer
• Raspberry Pi 4 (recommended)
• Raspberry Pi Camera Module
• Philips Hue bulbs + compatible Zigbee gateway (e.g. ConBee II)
• AWS account with an S3 bucket
• MQTT broker (e.g. Mosquitto or AWS IoT Core)

🚀 Usage

Below is an example of how to run the core scripts in sequence.

# 1️⃣ Test MQTT connectivity (optional)
python src/config/testMQTT.py

# 2️⃣ Run the camera module and upload images to AWS S3
python src/camera/cameraS3.py

# 3️⃣ Run the Human Activity Recognition model
python src/lightbulb/human_model.py

# 4️⃣ Compute Euclidean distances between humans and bulbs
python src/lightbulb/eulcidian_process.py

# 5️⃣ Execute the microcontroller logic to control the lights
python src/lightbulb/microcontroller_logic.py

You can also use the images in test_pictures/ to validate how the models and logic behave without running the camera live.

---

📊 Evaluation & Results

The system is evaluated based on:

• ✅ Detection performance for humans and bulbs
• ✅ Distance estimation quality using Euclidean distance
• ✅ Responsiveness of the lighting system
• ✅ Subjective user experience in terms of focus/relaxation

From experiments and state-of-the-art research:

• The underlying object-detection approach (YOLO-based) achieved high accuracy in detecting humans and light bulbs.
• The lighting control algorithm (distance-based) maintained a reliable mapping between human position and the correct bulb being activated.
• User-oriented tests and literature showed that using CCT ≈ 4300 K for work tasks supports sustained attention, while warmer scenes (around 2700 K) improved relaxation during breaks.

---

🧩 Repository Structure

WellOffice/
├── .vscode/
│   └── settings.json
│
├── HAR_models/
│   ├── FullHARBest.pt
│   ├── FullHARLast.pt
│   └── NewBestHAR.pt
│
├── src/
│   ├── camera/
│   │   └── cameraS3.py
│   │
│   ├── config/
│   │   ├── datasetCleaner.py
│   │   ├── datasetIndexer.py
│   │   └── testMQTT.py
│   │
│   ├── lightbulb/
│   │   ├── bestHAR.pt
│   │   ├── config.py
│   │   ├── eulcidian_process.py
│   │   ├── human_model.py
│   │   ├── lightbulbs_model.py
│   │   ├── lightbulbslogic.json
│   │   └── microcontroller_logic.py
│
├── test_pictures/
│   ├── blankimage.jpg
│   ├── chuanle1computer.jpeg
│   ├── chuanledevouring.jpeg
│   ├── chuanlejanicecomputer.jpeg
│   ├── chuanlelights1.jpeg
│   ├── chuanlelights2.jpeg
│   ├── chuanleworkingtest.jpeg
│   ├── gustavolights.jpeg
│   ├── gustavoworking.jpeg
│   ├── human_model_result.jpg
│   ├── janice1computer.jpeg
│   ├── janice2computer.jpeg
│   ├── janicedoingnothing.jpeg
│   ├── janicelights.jpeg
│   ├── lightbulb1on.jpeg
│   ├── lightbulb2on.jpeg
│   ├── lightbulb3on.jpeg
│   ├── lightsoff.jpeg
│   └── lightson.jpeg
│
├── .gitattributes
└── README.md

---

📬 Contact

Janice Sakina Akiza Nizigama

🔗 LinkedIn: https://www.linkedin.com/in/janice-nizigama

💻 GitHub: https://github.com/JSanizi

Tew Chuan Le

🔗 LinkedIn: https://www.linkedin.com/in/chuan-le-tew-0209b4203/

Eugenio Turcott Estrada

🔗 LinkedIn: https://www.linkedin.com/in/eugenio-turcott/

💻 Github: https://github.com/eugenio-turcott

Gustavo Lucio Sepúlveda

🔗 LinkedIn: https://www.linkedin.com/in/gustavo-lucio-a65944259/

💻 Github: https://github.com/gusls02

---

🧾 License

© 2024 Aalborg University — Academic / Non-Commercial use permitted with citation. For other uses, please contact the authors.