Thermal-infrared UAV tracking
Detection, tracking, and interactive visualization
Spring 2025 · course project
A UAV detection and tracking pipeline for video files and live streams, with a Streamlit interface and visual outputs.
ROLE
Course project author
APPROACH
YOLOv12 · BoT-SORT · Streamlit
Research question
Thermal UAV tracking connects frame-level detection with consistent target identities over time. I developed an interactive application that brings together detection, association, parameter control, and visual inspection of tracking outputs, supporting both video-file and live-input workflows.
My contribution
- Integrated YOLOv12 and BoT-SORT into a Streamlit application with single-target and multi-target configuration.
- Implemented video/frame input preparation, inference subprocess orchestration, and annotated output export.
- Exposed confidence, IoU, and ReID settings for interactive inspection of tracking behavior.
Technical approach
- Prepare video or frame inputs and run YOLOv12 detection, then associate detections across frames with BoT-SORT and optional ReID.
- Expose single-target and multi-target settings, confidence thresholds, and IoU parameters through Streamlit, with inference handled by a task runner.
- Render target boxes and tracks and export annotated video. The live-input workflow also includes a lightweight IoU association option.
Results & outcomes
- Delivered an application workflow spanning input preparation, parameter configuration, tracking execution, and output export.
- The project integrates model inference with task orchestration and visual review, making tracking experiments accessible through an interactive interface.
