Use Case Title :
Litter Detection in Open Area
Team Members :
Ramesh Choudhary,Keyur Bidawat
Use Case Importance :
Littering in public spaces harms the environment and affects community health and aesthetics. Our solution automates the detection of litter in open areas, enabling faster cleanup and better environmental monitoring.
Data Collection and Annotation
We captured 600 images in parks, roadsides, and open fields using a mobile camera in various lighting conditions.
Annotated classes: “Litter” and “Clean Area”
Annotation Tool: Roboflow
Total Images: 600
Model Training and Validation :
Model Used: YOLOv8-Nano
Metrics Monitored: mAP@0.5, precision, and recall
We started with 450 training and 150 validation images. After observing low recall on “Litter” detection, we added 100 more samples focusing on varied backgrounds and retrained the model.
Model Deployment and Demo Video :
The model achieved real-time performance at ~25 FPS with an accuracy of 87% on mobile using the YOLOv8 app.
We deployed a single YOLOv8-Nano model for inference.
Demo Video :
https://drive.google.com/file/d/1MOJzQAxMT_26Wiw_hG9LhlltXBS2gom2/view?usp=drivesdk
Conclusion :
Our model achieved strong real-time performance and reliable detection in diverse environments. Key learnings include the importance of background variation and the impact of lighting and object size on accuracy.
