Smart Space Occupancy Sign

An Intelligent Multi-modal Sensing System for Real-time Space Management

1. Project Overview

This project addresses the conflict between "Displayed Status" and "Actual Usage" in shared spaces (classrooms, meeting rooms, labs). We developed an intelligent occupancy sign that uses Multi-modal Heterogeneous Fusion to detect human presence with high reliability, even in "static" or "silent" scenarios.

2. System Architecture

The system adopts a distributed control architecture:

• Audio AI Module (ESP32): Captures acoustic features and runs a Random Forest classifier to identify human activities (talking, typing, moving).
• Microwave Radar Module (Arduino Uno): Detects micro-movements (breathing, slight shifting) via 24GHz microwave Doppler effect.
• Indicator System: A tri-color LED system (Red: Occupied, Yellow: Leaving soon, Green: Vacant).

3. Key Technical Highlights

3.1 AI Audio Feature Engineering

To overcome the limitations of far-field sound pickup, we implemented custom non-linear gain compensation:

• RMS Factor (*5.0): Amplifies the root-mean-square energy of the signal to capture faint distant sounds.
• Burst Factor (*4.0): Enhances transient pulse detection (e.g., pen clicking, page turning).
• Classifier: A Random Forest model optimized for low-power edge deployment on ESP32.

3.2 Heterogeneous Logic Fusion

We designed an Asynchronous Trigger-and-Hold logic to handle the "Blind Zone" and "Static Presence" issues:

1. Fast Wake-up: The system triggers "Occupied" if either the Radar or AI detects a signal within a 6-second window.
2. Robust Maintenance: A 30-minute cooldown timer prevents frequent flickering. The status only reverts to "Vacant" if no signals are detected by both sensors for the entire duration.

4. Repository Structure

├── ESP32-Audio-AI/          # Source code for AI inference & sound processing
├── Arduino-Uno-Radar/       # Radar signal processing & LED control logic
└── Hardware/            # 3D modeling files (.STL & .STEP) for the enclosure

5. Hardware Requirements

• Microcontrollers: ESP32-DevKitC, Arduino Uno R3
• Sensors: INMP441 (I2S Digital Microphone), LD2402 (24GHz Microwave Radar)
• Outputs: RGB LED Strip / Matrix
• Enclosure: Custom 3D-printed PLA housing

6. Reflection & Future Work

• Current Limitation: In extreme silence (e.g., deep sleep without movement), detection remains a challenge.
• Optimization: Future iterations will migrate to a single-chip solution (ESP32-S3) to reduce power consumption and latency.

7. Authors & Acknowledgments

Developed by a team of students from Shanghai Jiao Tong University (SJTU):

• Wang Yanzhe (Project Lead & AI Developer)
• Xiong Jinyi(AI Developer), Li Zeyu(3D-Model Designer), Pan Jingyuan(Electronic Control), Dong Yunhe(Report and PPT Making).

Special thanks to SJTU teachers:Yang Li and Wang Zechen for providing instructions, testing environment and equipment.