🎥 Camera Movement Detection

Ana Sayfa Live Demo GitHub Repo

Project Summary

This project is developed to detect camera motion in video or image sequences from a static camera. The goal is to determine only the physical changes of the camera (e.g., pan, tilt, shift, shake) independently of object movements in the scene.

  • Motivation: In video surveillance systems, detecting camera movement (e.g. tampering) is often more critical than detecting object motion.
  • Solution: Global motion was calculated using ORB feature matching, Optical Flow, and Affine methods between consecutive frames.
  • Detection: Translation values derived from the transformations are used to decide if there is "significant movement."

Technologies Used

Python Pillow, NumPy OpenCV Streamlit Docker Pytest SMTP
Dataset: Hugging Face syCen/CameraBench

Algorithm Descriptions

🔹 ORB + Homography

  • Purpose: Detect global camera movement by identifying and matching keypoints between consecutive frames.
  • Method:
    • Use ORB (Oriented FAST and Rotated BRIEF) to extract prominent keypoints.
    • Match keypoints between frames.
    • Analyze the homography matrix to determine if the motion is from the camera or objects.
  • Pro: Focuses on camera movement more than local object motion.
  • Con: Can be sensitive to dominant local motion.

🔹 Optical Flow

  • Purpose:Track per-pixel motion to estimate direction and intensity of movement.
  • Method:
    • Compute optical flow between two frames using Farneback method.
    • Calculate mean flow magnitude and variance.
    • High mean + low variance = likely camera movement.
  • Advantage:Filters out local object motion like hand/arm using variance.
  • Limitation: May miss very small camera shakes..

🔹 Affine + Good Features

  • Purpose: Model the global scene transformation (translation, rotation, scaling) to estimate camera movement.
  • Method:
    • Detect strong corners using Harris or Shi-Tomasi.
    • Track these corners across frames.
    • Analyze the affine transformation matrix.
  • Advantage: Sensitive to scene geometry.
  • Limitation: Unstable with too few corners.

Setup

1) Installation

git clone https://github.com/Arifkarakilic/camera-movement-detector.git
cd camera-movement-detector
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt

2) Real‑Time Email Alerts (SMTP)

Note: "Real‑Time Camera" works only locally; Streamlit Cloud doesn’t allow live camera.

# .env\ nSMTP_USER=youremail@gmail.com
SMTP_PASSWORD=your_app_password
RECEIVER_EMAIL=targetmail@gmail.com

Live detection triggers an alert email using a background threading worker so the UI stream remains smooth.

3) Launch

streamlit run camera-movement-detection/app.py

4) Usage

  • Select from built‑in frame folders or upload .mp4/.gif
  • Tune sensitivity with sliders
  • Inspect detected frames visually
  • Enable real‑time alerts to receive emails on movement

5) Testing

pytest tests/test_file_utils.py

Project Structure

camera-movement-detection/
├─ app.py
├─ config.py
├─ detectors/
│  ├─ movement_detector.py
│  ├─ movement_detector_affine.py
│  └─ movement_detector_optical.py
├─ logic/
│  ├─ realtime_detector.py
│  └─ video_processor.py
├─ ui/
│  ├─ sidebar.py
│  ├─ realtime_view.py
│  └─ video_analysis_view.py
├─ utils/
│  ├─ file_utils.py
│  ├─ time_utils.py
│  ├─ visual_utils.py
│  ├─ notify.py
│  └─ send_mail.py
├─ tests/
│  ├─ test_file_utils.py
│  └─ test_video_processor.py
├─ .streamlit/
│  └─ config.toml
├─ .env
├─ requirements.txt
├─ Dockerfile
└─ README.md

Sample Output

Detected camera motion frames: [13, 14, 15, 30]
Detected frames visualization
Detected motion frames visualized in the UI
SMTP sample output
SMTP alert sample

Run with Docker

With Dockerfile

docker build -t camera-app .
docker run -p 8501:8501 --env-file .env camera-app

Remember to list .env in .dockerignore.

AI Assistance & References

Some parts of this project were assisted and optimized using AI (OpenAI / ChatGPT).