Design of an autonomous litter detection and collection system for Icelandic beaches

• Main Author: Alain René Frey 1990-
• Other Authors: Háskólinn í Reykjavík
• Format: Master Thesis
• Language: English
• Published: 2024
• Subjects: Orkuverkfræði; Meistaraprófsritgerðir; Strandsvæði; Drónar; Sorp; Sustainable energy engineering; Seashore; Drone aircraft; Iceland
• Online Access: http://hdl.handle.net/1946/47664

Litter has become a large-scale problem in today’s world, exacerbated by the widespread use of single-use plastics and the expansion of the fishing industry. Countries like Iceland, with its extensive 5000km coastline, face year-round marine pollution challenges. Current efforts are inadequate due to their reliance on manual intervention for monitoring and cleanup. This thesis explores the feasibility of autonomous beach litter detection and collection using consumer drones, focusing on a mostly automated process for image analysis and geolocation. The proposed solution consists of drones autonomously scanning the beach, transmitting video streams and telemetry to a base station. A YOLOv8 detection model identifies trash objects in the video frames, and the location of these objects is calculated relative to the drone's position and stored in a database. This data can be used to dispatch autonomous pickup drones based on the shape and weight of the objects. The study focuses on two main components: the detection module and the location module. A YOLOv8 model was trained using a custom dataset of images from four different Icelandic beaches combined with preexisting datasets, achieving a mean average precision (mAP) of 0.78 and a precision of 83% at a confidence threshold of 0.8. The location module demonstrated reliable geolocation well within a 5x5m frame, necessary for subsequent trash pickup operations. Field surveys conducted on three Icelandic beaches validated the integrated system, showing its ability to detect and accurately geolocate trash. Environmental conditions such as ice presence, different beach types, and difficult lighting conditions negatively impacted the performance. The system's adaptability and scalability demonstrate its potential for large-scale deployment and continuous improvement through modular design. This work lays the foundation for practical, automated trash collection, contributing to environmental cleanup efforts with minimal human intervention.