VibES: Induced Vibration for Persistent Event-Based Sensing (3DV 2025)

VibES is a lightweight framework that enables event cameras to sense persistently in static or low-motion scenes by inducing controlled mechanical vibrations. Using a simple rotating unbalanced mass, VibES generates continuous event streams and compensates for the induced motion online and in real time, producing clean, motion-corrected data suitable for downstream perception tasks.

Why VibES? Event cameras are inherently motion-driven. VibES introduces a minimal, energy-efficient source of motion—paired with principled signal estimation—to unlock their use in otherwise static environments.

---

🚀 Key Features

• Persistent Event Generation Enables sensing in static scenes without mirrors, pan–tilt units, or complex optical setups.

• Ultra-Low Power Requires only 0.282 W using a small DC motor—orders of magnitude lower than robotic actuators.

• Online Motion Compensation Estimates and removes induced vibrations in real time without prior calibration or knowledge of physical parameters.

• High-Frequency Signal Estimation Robust frequency detection over 30–500 rad/s using Non-Uniform FFT (NUFFT) and tracking with Iterative EKFs (IEKF).

• Multi-Harmonic Support Accurately models complex vibration profiles beyond a single sinusoid.

• Proven Performance Gains Improves image reconstruction (NIQE ↓ 41%) and edge detection (contour length ↑ 87%).

• Broad Applicability Benefits edge detection, E2VID image reconstruction, feature tracking, and relative depth estimation via vibration-induced parallax.

---

🧩 System Overview

1. Mechanical Excitation: A rotating unbalanced mass induces small-amplitude, periodic motion of the event camera.
2. Event Stream Analysis: The induced motion produces a persistent stream of events, even in static scenes.
3. Frequency Estimation: NUFFT identifies dominant vibration frequencies from non-uniformly sampled events.
4. State Tracking: IEKFs track phase, amplitude, and offsets online.
5. Motion Compensation: Estimated motion is removed to yield stabilized, task-ready event data.

---

🛠️ Installation

Local Setup

1. Clone the Repository

git clone [email protected]:utiasSTARS/VibES.git --branch vibes
cd VibES
git submodule update --init --recursive

2. System Dependencies

Ensure FFTW3 and the Metavision SDK or OpenEB are installed

sudo apt-get update
sudo apt-get install -y libfftw3-dev libeigen3-dev libgflags2 libgflags-dev

3. Build (C++20)

mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j$(nproc)

Docker (Recommended)

Run with X11 forwarding for visualization, mount the dataset directory:

docker run --net=host --rm -v /PATH_TO/vibes_dataset/:/datasets --privileged -e DISPLAY=$DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix -it viciopoli/vibes:latest 

If you /bin/bash into the container different executables can be run. Example:

./compensation -c /datasets/intrinsics.json -i /datasets/data/logo_vib.hdf5 -o ../results/logo/ --tracker-x 557 --tracker-y 242

Press o to enable the overlay visualization.

---

💻 Usage

Core Executables

• compensation: Real-time vibration estimation and motion rectification.
• multi_tracker: Multi-feature tracking using the HASTE backend.
• frequency_estimation: Offline vibration frequency analysis.

Running the Compensation Tool

The primary tool is compensation. It requires an initial feature location (x, y) to begin tracking the vibration.

Basic Syntax:

./compensation -c <intrinsics.json> -i <file.hdf5> --tracker-x <x> --tracker-y <y>

If -i is not provided, the program will attempt to connect to a live camera.

---

📂 Dataset Download

To reproduce the results or run the examples, download the dataset from Hugging Face.

Using Git (Recommended)

Ensure you have git-lfs installed.

# Install Git LFS if needed
sudo apt-get install git-lfs
git lfs install

# Clone the dataset
git clone https://huggingface.co/datasets/viciopoli/VibES vibes_dataset

Files can also be downloaded manually from the Hugging Face.

---

📚 Citation

If you use VibES in your research, please cite:

@inproceedings{polizzi_2025_vibes,
  title     = {VibES: Induced Vibration for Persistent Event-Based Sensing},
  author    = {Polizzi, Vincenzo and Yang, Stephen and Clark, Quentin and
               Kelly, Jonathan and Gilitschenski, Igor and Lindell, David B.},
  booktitle = {International Conference on 3D Vision (3DV)},
  year      = {2025}
}

---

🤝 Acknowledgments

• FINUFFT – Non-uniform FFT computations
• HASTE – Event-based feature tracking
• Metavision SDK – Prophesee event camera interfacing

---

📎 Links

• 📄 Paper: https://arxiv.org/abs/2508.19094
• 🌐 Project Page: https://papers.starslab.ca/vibes/