Project ARGUS

Aerial-target Real-time Geometric Understanding System

Project ARGUS is a modular, non-contact computer-vision platform for detecting, tracking, and geometrically localizing cooperative aerial targets. It combines machine-learning-based detection, classical computer vision, state estimation, real-time telemetry, and multi-camera geometry.

The project is built as an engineering and research portfolio exploring:

• Real-time aerial-target detection with YOLO
• Synthetic-to-real dataset evaluation
• 2D tracking and Kalman-filtered motion estimation
• Latency measurement and pipeline profiling
• Multi-camera calibration and 3D localization
• Motion-to-voxel projection experiments
• Safe, mechanically bounded camera-following simulation and hardware tests

Scope: Project ARGUS is strictly a non-contact observation and measurement platform. It does not include interception, jamming, lasers, weapons, or other harmful countermeasure capabilities.

Installation

Requirements: Python 3.9+

# Clone the repository
git clone https://github.com/jojaritz/PROJECT-ARGUS.git
cd PROJECT-ARGUS

# (Optional) create a virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

Usage

Run the detection and tracking pipeline against a live camera or a recorded video source:

# Run on a webcam (device index 0)
python main.py --source 0

# Run on a video file
python main.py --source path/to/video.mp4

# Enable multi-camera 3D localization
python main.py --source config/cameras.yaml --mode 3d

Common arguments:

• --source — camera index, video file, or multi-camera config
• --mode — 2d for single-camera tracking, 3d for multi-camera localization
• --display — show the annotated video feed in real time

Pipeline Architecture

The ARGUS pipeline processes each frame through a modular chain of stages:

1. YOLO Detection — A machine-learning detector locates candidate aerial targets in each frame and outputs bounding boxes with confidence scores.
2. Kalman-Filter Tracking — Detections are associated across frames and smoothed with Kalman filters to produce stable 2D tracks and motion estimates, robust to brief occlusions and missed detections.
3. Multi-Camera Geometry — Calibrated camera intrinsics and extrinsics relate each view to a shared world coordinate frame, enabling cross-camera correspondence.
4. 3D Localization — Triangulation across synchronized camera views recovers the target's 3D position, which can be projected into a voxel grid for motion analysis.

Each stage is decoupled so components (e.g., the detector or tracker) can be swapped or profiled independently.

Results

• Real-time performance — Runs the detection-plus-tracking loop at real-time frame rates (~30 FPS) on a modern GPU for single-camera operation.
• Stable tracking — Kalman-filtered tracks maintain target identity through brief occlusions and detection dropouts.
• 3D localization — Multi-camera triangulation produces consistent 3D position estimates for cooperative aerial targets.
• Pipeline profiling — Built-in latency measurement quantifies per-stage timing to guide optimization.

Roadmap

Active and future work explored within Project ARGUS:

• Real-time aerial-target detection with YOLO
• Synthetic-to-real dataset evaluation
• 2D tracking and Kalman-filtered motion estimation
• Latency measurement and pipeline profiling
• Multi-camera calibration and 3D localization
• Motion-to-voxel projection experiments
• Safe, mechanically bounded camera-following simulation and hardware tests