🎥 AI-Powered Video Analytics System 🚀

An industry-ready, modular system for real-time object detection, multi-object tracking, and threat analytics powered by YOLOv8 and Deep SORT — with a beautiful and interactive Streamlit Dashboard.

---

🔍 Features

• ✅ YOLOv8-based Object Detection
• 🧠 Deep SORT Tracker with Appearance Features
• 🔁 Real-time Frame-by-Frame Processing
• 📊 Streamlit Dashboard with Upload, Preview, and Download
• 🚨 Custom Alert Zones to trigger visual warnings
• 🛠️ Modular, Scalable Codebase suitable for production

---

🛠️ Setup

1️⃣ Clone the Repo

git clone https://github.com/DIPESHGOEL27/video-analytics.git
cd video-analytics

2️⃣ Create & Activate a Virtual Environment

python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate

3️⃣ Install Dependencies

pip install -r requirements.txt

⚠️ Ensure torch is installed according to your GPU/CPU setup:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

4️⃣ Download Pretrained Models

• YOLOv8 weights
• mars-small128.pb appearance model in:

  app/tracking/deep_sort/resources/networks/mars-small128.pb
  

---

🚀 Run the App

streamlit run main.py

Upload any video, track objects live, and download processed output with overlays.

---

⚙️ Configuration (config.yaml)

video_input: "data/input_video.mp4"
video_output: "output/output_video.mp4"

yolo:
  model_path: "app/models/yolov8n.pt"
  confidence_threshold: 0.5

deep_sort:
  model_path: "app/tracking/deep_sort/resources/networks/mars-small128.pb"
  max_age: 30
  n_init: 3
  max_cosine_distance: 0.4
  nn_budget: 100

alert:
  enabled: true
  alert_zone: [100, 100, 400, 400]

---

📦 Deployment

To deploy this system:

• Host on Streamlit Cloud, Render, or Docker.
• Upload only .py files and model weights (not venv or cache folders).
• Ensure .gitignore includes __pycache__/, *.pyc, venv/.

---

👨‍💻 Developed by

Dipesh Goel
Third-year undergrad @ IIT Kharagpur
💼 Ocean Engineering and Naval Architecture
🌐 LinkedIn • ✉️ [email protected]

---

📃 License

This project is licensed under the MIT License.

---