How mwasifanwar/aerovision-ai works
Project Type and Purpose
• AeroVision-AI is a specialized computer-vision pipeline with a Streamlit-based frontend. Its purpose is real-time vehicle detection, tracking, and analytics on drone footage using YOLOv8 and OpenCV.
File Structure Analysis
• Top level:
• README.md → high-level description
• requirements.txt → Python dependencies
• yolov8n.pt → pre-trained YOLOv8 model
• outputs/ → generated CSVs & processed videos
• reports/ → (placeholder for static reports or summaries)
• src/ → core processing modules
• streamlit_app/ → web UI entry point
• src/ modules indicate a clear separation of concerns:
• detector.py → loads the YOLO model and runs inference
• tracker.py → associates detections across frames (e.g., Deep SORT or custom)
README Insights
• Technologies: Python, YOLOv8, OpenCV, Streamlit
• Features: live inference visualization, CSV analytics, dynamic watermarking, video export
Main Components and Interactions
a. Components
• Streamlit Frontend (streamlit_app/app.py)
• Inference Engine (src/detector.py with yolov8n.pt)
• Tracking Module (src/tracker.py)
• Visualization & Export (src/viz.py)
• Storage (local file system: inputs, outputs/*.csv, outputs/*.mp4)
b. Data Flows
1) User uploads or streams drone footage via Streamlit UI
2) Frontend calls Detector → processes frames → returns raw detections
3) Tracker ingests detections → produces tracked object IDs
5) Detection & tracking metadata → written to CSV files under outputs/
6) UI reads CSVs for analytics dashboards or downloads
c. Architectural Patterns
• Modular monolith: clear separation into detector / tracker / viz layers
• Pipeline pattern: sequential data processing stages
• MVC-like split between Streamlit “view/controller” and src “model”
d. Key Technologies
• YOLOv8 (Torch-based object detection)
• OpenCV (frame I/O, drawing, video encoding)
• Streamlit (rapid web UI for live feed and charts)