🚀 Real-Time Person & Vehicle Detection using YOLOv8 + OpenCV

This project uses Ultralytics YOLOv8 and OpenCV to perform real-time object detection on videos or live webcam feeds. It identifies and counts people and vehicles (cars, trucks, buses, motorcycles), displaying detection results and FPS directly on the video stream.

---

🧠 Features

• ✅ Real-time detection using YOLOv8 Nano (fastest variant)
• 🚗 Detects and counts people and vehicles.
• ⚡ Runs on GPU (CUDA) for high-speed inference.
• 💾 Logs all detected labels with timestamps to detected_objects.txt.
• 📊 Displays FPS, people count, and vehicle count on the screen.

---

Image preview : https://github.com/piyushkadam96k/car-human/blob/main/demo.jpeg

🧰 Requirements

Make sure you have the following dependencies installed:

pip install ultralytics opencv-python torch torchvision

💡 Optional: Install CUDA-compatible PyTorch for GPU acceleration.
See: https://pytorch.org/get-started/locally/

---

📁 Project Structure

📂 yolo-detection/
│
├── detection.py              # Main detection script
├── detected_objects.txt      # Auto-generated log of detections
├── requirements.txt          # Python dependencies
└── README.md                 # Project documentation

---

⚙️ How It Works

1. The YOLOv8 Nano model (yolov8n.pt) is loaded.
2. The model is moved to GPU (model.to('cuda')) for faster inference.
3. The program opens a video file or webcam feed.
4. Each frame is processed and analyzed:
   • Bounding boxes and class labels are drawn.
   • Counts for people and vehicles are displayed.
   • Detected objects are logged with timestamps.
5. Press q to exit the detection window.

---

▶️ Usage

1. Clone or download this repository

   git clone https://github.com/piyushkadam96k/car-human
   cd yolo-detection

2. Run the detection script

   python detection.py

3. Input source

   • Replace "The Fate of the Furious ｜ Harpooning Dom's Car.mp4"
     with your own video path or set it to 0 for webcam:

     cap = cv2.VideoCapture(0)

4. Exit

   • Press q anytime to close the video window.

---

🧾 Output Example

While running, the script shows:

People Count: 3
Vehicle Count: 5
FPS: 32.45

And logs detections like this in detected_objects.txt:

2025-10-30 12:05:43: person, car, truck
2025-10-30 12:05:44: person, motorcycle

---

⚡ Performance Tips

• For maximum FPS, use:
  • yolov8n.pt (Nano) or yolov8s.pt (Small)
  • Lower resolution frames (e.g., 640x400)
• Use GPU (CUDA) if available:

  model.to('cuda')

---

🧩 Model Details

Model	Size	Speed	mAP (50)	Best for
yolov8n.pt	6.2 MB	⚡ Fastest	~37.3	Real-time
yolov8s.pt	21.2 MB	Fast	~44.9	Better accuracy
yolov8m.pt	49.0 MB	Medium	~50.2	Balanced
yolov8l.pt	83.7 MB	Slower	~52.9	High accuracy

---

🧑‍💻 Author

Amit Kadam
📧 kadamamit462@gmail.com
📍 Bhalki
🔗 GitHub: piyushkadam96k

---

📜 License

This project is open-source and free to use under the MIT License