BugHunt

HexBug Multi‑Object Tracking Pipeline 🪲🎥

This project implements an automated multi-object tracking (MOT) pipeline for tracking HexBug heads across video frames. The pipeline takes a raw video as input and outputs a CSV file containing the head coordinates and persistent IDs for each detected HexBug in every frame.

Pipeline Overview

1. Video Preprocessing

   • The raw video is split into individual frames.
   • The aspect ratio of the first frame determines whether to use YOLOv8-S or YOLOv8-M for detection.

2. Detection (YOLOv8)

   • YOLOv8 is used to detect HexBug heads.
   • It outputs bounding boxes in the format: $x1</code>, <code>y1</code>, <code>x2</code>, <code>y2</code>, <code>confidence$.

3. Tracking (Modified SORT)

   • Detections are passed to a customized SORT tracker:

     • Bounding boxes are converted into Kalman filter states.
     • IDs are assigned via centroid-distance-based Hungarian matching with adaptive gating.
     • ID recycling and automatic ceiling enforcement are used to maintain consistent track counts.

4. Results

   • The tracker outputs a CSV file containing IDs and coordinates per frame: $frame</code>, <code>ID</code>, <code>x</code>, <code>y$.