SkySeer: Autonomous Night Sky Processing

AI-Powered Detection for Satellites, Meteors, and UAP

SkySeer uses Computer Vision to turn hours of raw night-sky footage into actionable data.
It autonomously isolates movement, calculates flight kinematics, and classifies objects without human supervision.

---

Technical Architecture

SkySeer operates on a custom 3-Stage Pipeline that transforms raw pixels into classified astronomical events.

1. Motion Isolation (Computer Vision)

We accept raw video (4K/1080p) and process it using Mixture of Gaussians (MOG2) background subtraction.

• Algorithm: cv2.createBackgroundSubtractorMOG2
• Parameters: Adaptive variance threshold (45), History (500 frames).
• Noise Reduction: Morphological Opening/Closing to remove sensor grain.

The pipeline filters 99% of empty sky frames, passing only active contours to the next stage.

2. Kinematic Feature Extraction

Every tracked object is transformed into an 11-Dimensional Feature Vector. We don't just look at the image; we look at the physics of the flight path.

Core Features	Visualized Data
• Velocity: Avg speed ($px/frame$) and Acceleration. • Linearity ($R^2$): How straight the trajectory is. • Consistency: Standard deviation of speed. • Blinking Score: Detects periodic flashing (planes). • Duration: Total time on screen.	Linear Regression analysis calculating the RMSE of a satellite trajectory ($R^2 \approx 0.99$).

3. Unsupervised Classification (Machine Learning)

Instead of relying on labeled datasets (which are scarce for night sky objects), SkySeer uses K-Means Clustering to separate objects based on their kinematic signatures.

• Scaling: All 11 features are normalized using StandardScaler.
• Clustering: Objects group naturally into two distinct classes:
  • Satellites: High Linearity, Low Speed Variance, Long Duration.
  • Meteors: High Velocity, High Burst Brightness, Short Duration.

Feature Distribution Analysis: Notice the clear separation in "Duration" and "Speed" between Satellites (Dark Blue) and Noise (Light Blue).

---

Detection Results

The system outputs a processed video with color-coded bounding boxes and valid confidence scores.

🔴 Satellite Detection	🟡 Meteor Detection
Class 0 Characteristics: • High Linearity ($R^2 &gt; 0.95$) • Constant Velocity • Long Duration (>3s)	Class 1 Characteristics: • High Velocity Spike • Transient Duration (<2s) • Brightness Flare

---

Data Visualization

SkySeer provides an interactive dashboard to analyze the night's traffic.

Speed Heatmaps	Classification Breakdown
Visualizing traffic lanes and velocity hotspots.	Quantifying the ratio of Satellites vs. Junk/Noise.

---

Configuration & Output

Fine-Tuning Controls	Data Export
	The Output Package: SkySeer generates a comprehensive mission folder containing: • Satellite_detections.mp4 (10x Speed) • analysis_report.csv (Scientific Data) • SUMMARY.txt (Quick Stats)

---

Project Structure SkySeer/ ├── src/ │ ├── motion_engine.py # Core MOG2 Logic │ ├── kinematics.py # Velocity & Linearity Math │ ├── classifier.py # Scikit-Learn K-Means Logic │ └── app.py # Streamlit Entry Point ├── Screenshots/ # Documentation Images ├── requirements.txt # Dependencies └── README.md # Documentation

Author Iker Simarro Cuevas

Focus: Computer Vision, Signal Processing, Scientific ML

https://www.linkedin.com/in/iker-simarro-546169227/

---

Installation & Usage

1. Clone the Repo

git clone [https://github.com/IkerSimarro/SkySeer.git](https://github.com/IkerSimarro/SkySeer.git)
cd SkySeer