A powerful GUI-based tool for video object detection, tracking, and structured data extraction using YOLO (Ultralytics).
Designed for traffic analysis, pedestrian behavior research, and dataset generation, this tool converts raw videos into clean, analysis-ready CSV data.
- 🎯 YOLO-based object detection (Ultralytics)
- 🧠 Multi-object tracking (ByteTrack)
- 📊 Automatic CSV data extraction
- 🎥 Live detection preview
- 🧾 Annotated video export
- 📁 Multi-video queue system
- ⚡ GPU acceleration (CUDA support)
- 🎛️ Full GUI (Tkinter-based)
- 🔍 Class filtering (by name or ID)
- 🚀 Batch processing for high performance
Input Video → Detection → Tracking → Data Extraction → CSV Output
Each detected object is exported with:
| Field | Description |
|---|---|
| video_name | Source video |
| frame | Frame index |
| time_sec | Timestamp |
| track_id | Unique ID (if tracking enabled) |
| class | Object class |
| confidence | Detection confidence |
| x1_rel | Normalized bounding box |
| y1_rel | Normalized bounding box |
| x2_rel | Normalized bounding box |
| y2_rel | Normalized bounding box |
git clone https://github.com/trewto/YOLO-Video-Analytics-Extractor.git
cd YOLO-Video-Analytics-Extractorpip install -r requirements.txtDependencies used in this project include:
Run the GUI:
python yolodetectionextractor_v13.py
| Option | Description |
|---|---|
| Tracking | Enables ByteTrack tracking |
| Live View | Real-time visualization |
| Export CSV | Save detections |
| Full GPU | Maximum speed (disables live & tracking) |
| Frame Skip | Process every Nth frame |
| Batch Size | Controls GPU load |
| Resize | Optimize processing speed |
| Class Filter | Select specific objects |
- YOLO model loads once for efficiency
- Frames processed in batches
- Optional tracking using ByteTrack
- Results converted to normalized coordinates
- CSV written incrementally (chunked) for large videos
- Tracking forces batch size = 1
- GPU strongly recommended for performance
- Large videos use chunked CSV writing to prevent memory issues