Sentinel Fall Detector

A real-time fall detection system for the Raspberry Pi 5 that uses an LSM6DSO32 IMU sensor to distinguish genuine falls from everyday movements like sitting down. It combines Kalman-filtered sensor fusion, a multi-stage fall detection state machine, and FFT-based gait analysis to deliver accurate alerts with a modern Qt Quick dashboard.

---

Table of Contents

• About The Project
• Getting Started
• Usage
• Roadmap
• License
• Contact

---

About The Project

Motivation

Consumer fall detection solutions are often opaque, expensive, or locked into proprietary ecosystems. Sentinel Fall Detector is an open, transparent implementation built on commodity hardware that demonstrates how multi-stage signal processing can separate real falls from false positives — a problem that plagues simpler threshold-based detectors. It serves as both a functional prototype and a reference architecture for IMU-based safety systems.

Key Features

• Kalman-Filtered Orientation Tracking — A 2-state Kalman filter fuses accelerometer and gyroscope data to produce stable roll and pitch estimates that resist noise and drift.
• Multi-Stage Fall Detection — A three-phase state machine (freefall detection, impact detection, decision matrix) classifies events as hard falls, safe sitting, posture shifts, or benign impacts by comparing current state against a 500ms historical buffer.
• Predictive Gait Analysis — A radix-2 FFT analyzes 128-sample windows of acceleration magnitude, extracting gait energy (0.5–3.5 Hz) and tremor energy (4–15 Hz) to compute a fall risk score before a fall occurs.
• Real-Time 3D Dashboard — A Qt Quick UI renders a 3D device model that mirrors physical orientation in real time, alongside a live force meter, status indicators, and fall risk warnings.
• False Positive Rejection — The decision matrix uses relative posture change and vertical velocity to distinguish controlled movements (sitting, stumbling) from genuine falls, preventing false alarms.

Built With

---

Getting Started

Prerequisites

Requirement	Details
Hardware	Raspberry Pi 5 with display, LSM6DSO32 IMU sensor via Qwiic Shim (I2C address 0x6A)
OS	Raspberry Pi OS or any Linux with I2C support
Compiler	GCC or Clang with C++17 support
CMake	Version 3.16 or higher
Qt 6	Modules: Quick, Core, Gui
I2C Tools	i2c-tools package (for diagnostics)

Installation

1. Enable I2C on the Raspberry Pi (if not already enabled):

   sudo raspi-config
   # Interface Options → I2C → Enable

2. Install dependencies:

   sudo apt update
   sudo apt install -y cmake qt6-declarative-dev qt6-base-dev i2c-tools

3. Verify sensor connectivity:

   i2cdetect -y 1
   # Address 6a should appear in the grid

4. Clone and build:

   git clone <repository-url>
   cd fall-detection
   cmake -S . -B build
   cmake --build build -j$(nproc)

---

Usage

Running the Application

./build/FallDetector

If GPU drivers are not configured, use the software rendering backend:

QT_QUICK_BACKEND=software ./build/FallDetector

Remote Deployment to Raspberry Pi

scp -r ./* user@raspberrypi5:~/Documents/FallDetector/
ssh -t user@raspberrypi5 'cd ~/Documents/FallDetector && cmake -S . -B build && cmake --build build -j$(nproc) && export DISPLAY=:0 && ./build/FallDetector'

A preconfigured VSCode task (Deploy and Run on Pi) is also available in .vscode/tasks.json.

Dashboard States

State	Background	Status Text	Trigger
Monitoring	Deep Slate	Monitoring...	Default
Fall Detected	Crimson Red	EMERGENCY: FALL DETECTED	Freefall + high impact + posture change > 45°
Safe Descent	Deep Slate	Sitting Down / Safe Descent	Controlled downward motion
Fall Risk Warning	Amber pulse	HIGH FALL RISK DETECTED	Gait risk score > 0.6

Testing Scenarios

Detailed manual validation procedures are documented in TESTING_GUIDE.md, covering:

• Kalman filter orientation accuracy
• Hard fall detection and emergency alert
• False alarm rejection (stumbles, taps)
• Daily activity recognition (sitting down)
• Predictive gait analysis with FFT

---

Roadmap

• Kalman-filtered sensor fusion for roll/pitch estimation
• Multi-stage fall detection with decision matrix
• FFT-based predictive gait analysis
• Real-time Qt Quick dashboard with 3D visualization
• Automated unit and integration test suite
• CI/CD pipeline for cross-compilation targeting Raspberry Pi
• Bluetooth/Wi-Fi alert forwarding to caregiver devices

---

License

Distributed under the MIT License. See LICENSE.txt for more information.

---

Contact

Jason Tran

• Email: tran219jn@gmail.com
• Website: jasontran.pages.dev