Kalman-YOLO Improving YOLO Tracking Performance through the Integration of a Kalman Filter for a Beach Cleaning Robot

Abstract

Ocean waste poses a significant threat to both human and marine life as industries and individuals continue to dump garbage into the ocean.Sea creatures are poisoned by materials such as plastics and chemicals, which in turn contaminate humans who consume them.This paper introduces an innovative approach using Image Instance Segmentation with YOLOv8 to segment and track beach garbage.However, YOLOv8's object tracking struggles in dynamic environments with challenges like occlusion, shadows, and perspective changes in RGB frames.To address this, the author presents Kalman-YOLO, combining the Kalman Filter with YOLO for improved performance.Results show notable performance improvement, especially in tracking garbage for the Beach Cleaning Robot.