🧠 Video Object Tracking with YOLOv8 + DeepSORT

This project implements a real-time video object tracking system using YOLOv8 for detection and DeepSORT for tracking. It supports training on MOT datasets, evaluating performance, and running on custom videos to track and identify pedestrians (or other objects).

---

📂 Project Structure

video_object_tracking/
├── data/
│ ├── MOT_YOLO/ # Preprocessed MOT dataset (images + labels)
│ ├── pedestrian.mp4 # Example input video
├── evaluation/
│ └── evaluate.py # Evaluation script
├── models/
│ ├── deep_sort_tracker.py
│ ├── trainer.py
│ └── yolo_seg_detector.py
├── outputs/ # Results: tracked videos, .txt outputs
├── runs/ # YOLO training + detection outputs
├── utils/
│ └── main.py # Utility functions
├── track_predict.py # Run tracking on any video
├── train.py # Train YOLOv8 on MOT dataset
├── requirements.txt
└── README.md

---

🚀 Features

• 🔍 YOLOv8 for real-time object detection and segmentation
• 👣 DeepSORT for multi-object tracking
• 📊 Train on custom MOT17-style datasets
• 🎥 Track pedestrians (or other classes) in videos
• 🧪 Output results in MOT format for evaluation

---

🔧 Installation

1. Clone the repository:

git clone https://github.com/your-username/video_object_tracking.git
cd video_object_tracking

2. Install dependencies:

pip install -r requirements.txt

3. Download YOLOv8 weights:

wget https://github.com/ultralytics/assets/releases/download/v0.0.0/yolov8n.pt

---

🏋️ Train on MOT Dataset

Ensure the dataset is structured like this:

data/MOT_YOLO/
├── images/
│   ├── train/
│   └── test/
├── labels/
│   ├── train/
│   └── test/
└── config.yaml

To start training:

python train.py

Update model path (yolov8n.pt, yolov8s.pt) and training parameters in train.py.

---

🧪 Evaluate Trained Model

python evaluation/evaluate.py

Generates bounding box predictions in MOT format and evaluates them (optionally).

---

📹 Track Pedestrians in Video

Use your trained YOLOv8 model with DeepSORT on a video:

python track_predict.py \
  --video data/pedestrian.mp4 \
  --output_video outputs/output.mp4 \
  --output_txt outputs/output_mot.txt \
  --weights runs/train/yolov8n-mot17/weights/best.pt

This saves:

Annotated video with tracked objects

MOT-format .txt file with per-frame object positions

---

📌 Output Format (MOT)

Each row in output_mot.txt:

frame_id, track_id, x, y, w, h, 1, -1, -1, -1

---

🧠 Model Architecture

YOLOv8: For fast object detection or segmentation.

YOLOSegDetector: Wrapper over YOLOv8 for detection.

DeepSORT: Tracks objects across frames using motion + appearance.

DeepSortTracker: Wrapper class that handles integration with detections.

---

🛠 TODO

Add web UI for upload + tracking

Add detection class filters

---

🤝 Contributing

Open to suggestions, bug fixes, and feature contributions. PRs welcome!