📌 Real-Time Object Tracking with YOLOv12 and OpenCV

A real-time object tracking system that combines YOLOv12 detection with OpenCV's CSRT tracker. Point your webcam at any scene, click on an object, and the system tracks it continuously — automatically re-detecting it if tracking is lost.

---

🎯 How It Works

1. Detection phase — YOLOv12 runs on the live webcam feed and draws bounding boxes around all detected objects in real time.
2. Selection — Click on any detected object with your mouse. If multiple bounding boxes overlap at that point, the system picks the smallest (most specific) one.
3. Tracking phase — OpenCV's CSRT tracker takes over, tracking the selected object frame by frame without running the full detection model on every frame.
4. Re-detection — If the tracker loses the object (e.g., it leaves the frame or is occluded), YOLOv12 re-runs automatically and resumes tracking as soon as the object reappears with confidence ≥ 0.7.

---

✅ Features

• Real-time detection with YOLOv12 (nano variant — fast and lightweight)
• Click-to-track — select any object from the live feed with a single mouse click
• CSRT tracking — accurate, robust tracker for smooth object following
• Automatic re-detection — recovers tracking after object loss without user intervention
• Smallest-box selection — intelligently selects the most specific bounding box when objects overlap

---

🛠️ Tech Stack

Tool	Role
YOLOv12 (Ultralytics)	Real-time object detection
OpenCV CSRT Tracker	Frame-by-frame object tracking
Python 3.8+	Core language

---

📁 Project Structure

real-time-object-tracking/

│

├── object_tracker.py # Main script — detection, selection, and tracking

├── test_webcam.py # Simple webcam test to verify camera access

├── yolov12n.pt # YOLOv12 nano pretrained model weights

└── README.md

⚙️ Setup

1. Install dependencies:

pip install ultralytics opencv-python

2. Verify your webcam works:

python test_webcam.py

3. Run the tracker:

python object_tracker.py

---

🖱️ Controls

Action	Key / Input
Select object to track	Left mouse click on detected object
Quit	Press q

---

Demo Video

https://www.linkedin.com/feed/update/urn:li:activity:7300530083985780736/

📝 Notes

• The model used is yolov12n.pt (nano) — optimized for speed on CPU.
• CSRT is chosen over simpler trackers (KCF, MOSSE) for its higher accuracy on fast-moving or partially occluded objects.
• Re-detection only triggers if the CSRT tracker explicitly fails — keeping CPU usage low during stable tracking.