SpDly
Studios

SpDly Studios / Project documentation

OEUR — Wireless Occupancy Estimation via RSSI Signal Processing

Independent static recordProject archive

Version: 1.0

Status: Research Completed (Expandable)

Category: Research · Signal Processing · Wireless Sensing · Machine Learning · Communication Systems

Institution: Madras Institute of Technology, Anna University

Research Period: 2025 – 2026

Conference: Engineering Advances 2025

Research Type: Applied Wireless Signal Processing

Resume Reference: Developed an RSSI-based occupancy estimation system without dedicated sensors using statistical signal processing, achieving 89.47% classification accuracy.

Overview

Wireless Occupancy Estimation via RSSI Signal Processing (OEUR) is a device-free occupancy sensing system that estimates the number of people present in an indoor environment by analyzing fluctuations in Wi-Fi Received Signal Strength Indicator (RSSI) measurements between two ESP8266 devices.

Unlike camera-based or PIR-based solutions, the system requires no dedicated occupancy sensors, preserving privacy while remaining inexpensive and easy to deploy. The project combines wireless communication, digital signal processing, statistical analysis, and machine learning-inspired feature engineering to classify occupancy levels based solely on RSSI characteristics.

Motivation

Traditional occupancy estimation systems rely on cameras, PIR sensors, infrared arrays, pressure sensors, RFID, or computer vision — approaches that often suffer from privacy concerns, high deployment cost, additional hardware requirements, complex installation, poor scalability, and high computational requirements.

Since wireless signals naturally propagate through occupied spaces, human presence alters propagation characteristics through absorption, scattering, reflection, and multipath fading. These changes are measurable in RSSI values, enabling occupancy estimation using existing wireless infrastructure.

Objectives

  • Develop a completely sensorless occupancy estimation system using commodity Wi-Fi hardware (ESP8266).
  • Collect high-frequency RSSI measurements and design a robust statistical filtering pipeline.
  • Extract meaningful signal features and classify occupancy states with high accuracy.
  • Demonstrate feasibility without machine vision, minimize deployment cost, and preserve user privacy.

    • Problem Statement

    Indoor occupancy estimation is critical for smart buildings, HVAC optimization, energy conservation, security monitoring, space utilization, disaster management, and human activity monitoring. Most existing solutions require dedicated sensing hardware or invasive monitoring systems.

    The challenge is determining whether occupancy information can be inferred solely from changes in wireless signal strength while maintaining acceptable classification accuracy.

    Research Questions

  • How strongly does occupancy affect RSSI?
  • Which statistical features best separate occupancy classes?
  • How much does filtering improve classification?
  • What sampling frequency is required?
  • How stable is RSSI over long durations?
  • Which divergence metrics best distinguish distributions?
  • How robust is the system against environmental noise?

    • Background Theory

    RSSI

    Received Signal Strength Indicator measures received radio signal power. Factors affecting RSSI include distance, multipath propagation, reflection, refraction, human obstruction, antenna orientation, environmental interference, and hardware variability.

    Human Influence on RF Signals

    Human bodies contain significant water content, which absorbs electromagnetic energy. Human movement introduces dynamic multipath effects, causing measurable RSSI fluctuations. Primary propagation effects include absorption, reflection, diffraction, shadowing, and multipath fading — altering the statistical distribution of RSSI measurements.

    System Architecture

    
ESP8266 Transmitter

       │

       │ Wi-Fi Signal

       ▼

Indoor Environment (Humans influence propagation)

       │

       ▼

ESP8266 Receiver

       │

RSSI Sampling → Serial Transmission → Python Processing Pipeline

       │

Filtering → Feature Extraction → Classification → Occupancy Prediction

    Hardware

    Transmitter: ESP8266, constant packet transmission, fixed location, stable power source

    Receiver: ESP8266, continuous RSSI acquisition, serial communication with PC

    Host Computer: Responsible for data collection, processing, visualization, feature extraction, and classification

    Experimental Setup

    Occupancy Classes

  • Empty room (0 persons)
  • One person
  • Two people
  • Three people
  • Higher occupancy (if applicable)

    • Each class collected independently under similar environmental conditions.

    Signal Processing Pipeline

    
Raw RSSI

    ↓

Median Filter (window size: 9 samples)

    ↓

Exponential Moving Average (α = 0.2)

    ↓

Histogram Generation

    ↓

Probability Mass Function (PMF)

    ↓

Statistical Feature Extraction

    ↓

Classification → Occupancy Prediction

    Median Filtering

    Removes impulsive spikes and outliers while preserving trends. Window size: 9 samples.

    Exponential Moving Average

    Smooths remaining fluctuations. Alpha: 0.2. Retains gradual signal variations while suppressing random noise.

    Histogram Generation

    Converts RSSI values into histograms to capture signal distribution instead of individual samples, enabling robust statistical comparison.

    Probability Mass Function (PMF)

    Normalizes histogram values to represent RSSI distribution as probabilities, enabling comparison using information-theoretic metrics.

    Statistical Features Extracted

  • Mean
  • Standard deviation
  • Variance
  • Distribution shape
  • Probability density characteristics
  • Divergence measures

    • Divergence Metrics

    Jensen-Shannon Divergence

    Symmetric comparison of probability distributions. Stable, bounded, and robust.

    Kullback-Leibler Divergence

    Measures information loss between distributions. Useful for occupancy differentiation.

    Bhattacharyya Distance

    Measures overlap between probability distributions. Smaller overlap indicates better class separation.

    Software Stack

    Python · PyQt · NumPy · SciPy · Matplotlib · Serial Communication · Custom Processing Pipeline

    GUI Features

  • Live RSSI graph
  • Filtered signal display
  • Histogram visualization
  • PMF visualization
  • Divergence metrics readout
  • Occupancy prediction display
  • Recording controls

    • Performance

    Demonstrates that RSSI contains sufficient statistical information for practical occupancy estimation under controlled conditions.

    Advantages

  • No cameras required; privacy preserving
  • Low cost and non-invasive
  • Commodity Wi-Fi hardware (ESP8266)
  • Easy deployment and real-time capable
  • Expandable to multi-room and multi-channel setups

    • Limitations

  • Environment dependent; sensitive to furniture changes
  • Device calibration required
  • Wi-Fi interference and multipath variability
  • Limited generalization without retraining

    • Applications

    Smart classrooms · Smart offices · Energy optimization · Building automation · Occupancy analytics · Disaster shelters · Hospital monitoring · Conference rooms · Libraries · Public infrastructure

    Challenges Encountered

  • RSSI instability and environmental noise
  • Packet loss and hardware variability
  • Long-duration recording consistency
  • Statistical feature selection and class overlap
  • Signal smoothing without information loss

    • Lessons Learned

  • Raw RSSI is highly noisy; statistical distributions are more informative than individual measurements.
  • Proper filtering significantly improves classification performance.
  • Feature engineering is more important than increasing model complexity for this problem.
  • Consistent experimental procedures are essential for reproducible results.

    • Future Improvements

  • Multi-room occupancy estimation
  • Multi-channel RSSI fusion
  • CSI (Channel State Information) integration
  • Deep learning classifiers
  • Transfer learning across environments
  • Automatic calibration
  • Edge processing on embedded devices
  • Real-time cloud dashboard
  • Multi-floor deployments

    • Publications & Presentation

    Conference: Engineering Advances 2025 International Conference

    Paper: Wireless Occupancy Estimation via RSSI Signal Processing

    Presentation: Conference oral presentation describing methodology, experiments, signal processing pipeline, and achieved classification accuracy.

    Keywords

    RSSI · Occupancy Estimation · Device-Free Sensing · Wi-Fi Sensing · Indoor Localization · Statistical Signal Processing · ESP8266 · Feature Extraction · Jensen-Shannon Divergence · Kullback-Leibler Divergence · Bhattacharyya Distance · Probability Mass Function · Wireless Sensing · Human Detection · Embedded Systems · Python · PyQt · Digital Signal Processing