Smart Surveillance with YOLOv8 and DeepSORT

This project is a computer vision mini-project for real-time smart surveillance. It detects people with YOLOv8, tracks them with DeepSORT, estimates motion between frames, logs crowd statistics, and raises a simple crowd alert.

The repo is now organized so it can be shown on GitHub more cleanly:

• detection.ipynb: notebook walkthrough and quick experimentation
• run_surveillance.py: reusable script for real-time detection and tracking
• plot_people_log.py: plots people-count trends from the generated CSV log
• evaluate_predictions.py: optional evaluation script for labeled prediction CSVs
• evaluation_labels_template.csv: template for adding y_true and y_pred
• requirements.txt: Python dependencies
• archive/: local dataset folder used only during experimentation
• assets/: README visuals such as banner, demo frame, and workflow diagram

Academic Summary

A short academic-style abstract is available in PROJECT_ABSTRACT.md.

Project Showcase

Demo Frame

Workflow Diagram

Workflow

1. Run the surveillance pipeline on webcam or video input.
2. Save annotated output video and CSV logs.
3. Plot people-count trends from the log.
4. Optionally evaluate predictions with a labeled CSV file.

Features

• Person detection using YOLOv8
• Multi-object tracking using DeepSORT
• Current visible people count
• Total unique tracked people count
• Basic frame-difference motion estimation
• Crowd alert based on configurable threshold
• CSV logging for later analysis

Setup

Create and activate a Python 3.10 environment, then install dependencies:

pip install -r requirements.txt

Dataset Download

This project was tested with the Kaggle dataset below:

suryaprabhakaran2005/road-accidents-from-cctv-footages-dataset

You can download it with kagglehub using:

import kagglehub

# Download latest version
path = kagglehub.dataset_download(
    "suryaprabhakaran2005/road-accidents-from-cctv-footages-dataset"
)

print("Path to dataset files:", path)

The printed path is the local dataset folder on your machine, and it may differ from system to system.

Run the Project

Option 1: Notebook

jupyter notebook detection.ipynb

Option 2: Python script

Use webcam:

python run_surveillance.py --source 0

Use a video file:

python run_surveillance.py --source path/to/video.mp4 --no-display

This creates:

• output.avi: annotated output video
• log.txt: CSV log with frame, current people, total unique people, and motion level

Plot Results

python plot_people_log.py --log log.txt

To save the plot:

python plot_people_log.py --log log.txt --output people_plot.png

Optional Evaluation

This repository does not include a labeled validation file by default, so metrics are optional instead of fake.

To evaluate predictions:

1. Copy evaluation_labels_template.csv
2. Fill it with two columns: y_true and y_pred
3. Run:

python evaluate_predictions.py --csv evaluation_labels.csv

Notes for GitHub Submission

• Do not push the full dataset folder to GitHub.
• Do not push generated outputs unless you want a small sample demo.
• If you submit this academically, keep the project story consistent: this repo is a surveillance / people tracking project, not accident severity classification.
• Add 2-3 screenshots or a short GIF in the final GitHub version for stronger presentation.
• Add demo media inside the assets/ folder before final sharing.

GitHub Polish Added

• Reusable Python scripts for running, plotting, and evaluation
• Optional evaluation template instead of fake metrics
• GitHub issue templates and PR template
• Lightweight GitHub Actions syntax check

Suggested GitHub Push Flow

git init
git add .
git commit -m "Initial smart surveillance project"
git branch -M main
git remote add origin <your-github-repo-url>
git push -u origin main

Author

MAYANK PANDEY

License

This project is licensed under the MIT License. See the LICENSE file for details.