LitterEye – AI-Powered Littering Detection

AI system that detects littering using vision and action recognition, simulates penalties, and helps keep public spaces clean.

Comment

Inspiration

With rising urbanization, littering on streets and public spaces has become a growing issue, contributing to pollution and unhealthy environments. We wanted to explore how AI-powered vision systems could simulate a real-world solution for identifying littering behavior and enforcing penalties — without requiring custom hardware. Our goal was to merge computer vision + action recognition into a practical civic-use case.

What it does

LitterEye is an AI prototype that:

  • Detects trash items on the ground (plastic, glass, paper, etc.).
  • Recognizes when a person performs a “throwing” or “dropping” action.
  • Fuses both events to simulate a littering detection system.
  • Issues an automated penalty entry and logs it in a database.
  • Displays results in a simple Streamlit dashboard with charts, tables, and video playback.

How we built it

  • Datasets:

  • Tech Stack:

    • YOLOv8 for trash detection (with simulation fallback).
    • MediaPipe for action recognition.
    • Streamlit for building the interactive demo app.
    • SQLite for storing penalties and logs.

  • System Flow:

    1. Input video (uploaded or demo clip).
    2. Run trash detection + action recognition.
    3. If “trash detected” ∧ “throwing action detected” → flag as littering.
    4. Log penalty in database + visualize on dashboard.

Challenges we ran into

  • Finding a relevant dataset that combined human actions with littering scenarios.
  • Training models on limited compute without using custom hardware.
  • Designing a simulation pipeline that feels close to real-world usage.
  • Integrating different ML models (object detection + action recognition) into one smooth pipeline.

Accomplishments that we're proud of

  • Built a complete simulation prototype with no external hardware.
  • Successfully fused object detection and action recognition into a single decision system.
  • Delivered a working demo app with database, dashboard, and penalty system.
  • Created a reusable and extendable code structure.

What we learned

  • How to integrate multiple pre-trained ML models into one workflow.
  • Practical challenges of building civic AI solutions.
  • The importance of datasets, preprocessing, and simulation design when hardware isn’t available.
  • How to design a project that can scale from a prototype → real-world deployment.

What's next for LitterEye – AI-Powered Littering Detection

  • Deploying the system on edge devices (CCTV, Raspberry Pi, Jetson Nano) for real-time use.
  • Adding face anonymization + privacy-preserving tracking.
  • Expanding the dataset with real-world littering scenarios.
  • Collaborating with municipalities to pilot the system in smart cities.
  • Exploring reward-based systems (e.g., citizen credits for reporting littering) in addition to penalties.

Built With

Share this project:

Updates