Person Re-Identification and Counting with YOLO11n + BoT-SORT

This project integrate human detection & tracking models to detect the accurate number of people present in a scene. One big struggle of object trackers is that the trackers cannot remember whether an object has appeared before. This is solved using BoT-SORT's ReID procedure. This ensures the same person retains he/her ID when they reappear.

This project demonstrates real-time person detection, tracking, and re-identification (ReID) using YOLO and BoT-SORT tracker.

Unlike simple trackers that lose track of people during occlusion or camera movement, ReID ensures that the same person retains their ID even after disappearing and reappearing.

The system also counts the number of unique people who appear in the video.

---

✨ Features

• Person detection with YOLO11 models
• Re-identification (ReID) using BoT-SORT → keeps the same ID after occlusion
• Unique person counting → tracks how many distinct people appeared throughout the video
• Configurable frame-skipping for faster processing (process_every_n)
• Export annotated video with bounding boxes, IDs, and person count overlay

---

📊 Example Output

• Bounding boxes around detected people
• Each person assigned a unique persistent ID
• Total unique person count displayed on screen
• Final summary of total unique people at the end

---

🎬 Demo

correct.identification.after.getting.overlapping.by.another.person.mp4

Explanation of the Demo:
In the above video, the person with ID 5 got temporarily occluded by ID 1. However, as soon as ID 5 reappeared, the tracker correctly regained its original ID (5), proving that ReID successfully prevents ID switches after occlusion.

---

🛠️ Use Cases

This system can be adapted for a wide range of real-world applications:

1. Crowd Analytics – Count the number of unique visitors in shopping malls, stadiums, or public spaces.
2. Security & Surveillance – Track individuals across multiple cameras, even after temporary occlusion.
3. Event Monitoring – Estimate crowd size at concerts, rallies, or festivals.
4. Smart Retail – Understand customer foot traffic patterns without duplicate counts.
5. Workplace Safety – Monitor restricted zones to ensure only authorized personnel are present.

---

🚀 How It Works

1. YOLO Detection – Detects persons in each frame (class=0).
2. BoT-SORT Tracking – Assigns IDs and tracks movements.
3. ReID Embeddings – Ensures consistent IDs across occlusions.
4. Unique ID Tracking – Counts the number of distinct IDs that appeared.
5. Video Export – Saves an annotated video with IDs and statistics.

---

📦 Installation

git clone https://github.com/yourusername/person-reid-counter.git
cd person-reid-counter
pip install -r requirements.txt

---

▶️ Run

python main.py --video path/to/your/video.mp4

---

✅ Example Result

👥 Total unique persons detected: 12
✅ Output video saved as: output_video.mp4