Optimization of Handling Lost Objects Using the Multiple Object Tracking Algorithm and Ensemble Kalman Filter

This study addresses the difficulties in recognizing overlapping objects in digital images, which often hinder the separation and identification of each object, leading to reduced accuracy and efficiency in object recognition. To mitigate these issues, the study integrates the YOLOv5 algorithm with the Ensemble Kalman Filter (EnKF), a statistical method known for enhancing performance and improving prediction accuracy. Although YOLOv5 is well-regarded for its efficient real-time object detection, it struggles with partially covered or partially visible objects. The combined YOLOv5 and EnKF approach was tested on various datasets featuring partially occluded or partially visible objects. The results revealed significant improvements across key performance metrics: accuracy, recall, F1 Score, and Intersection over Union (IoU). YOLOv5 alone achieved an average accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{48.46 \%}$</tex>, which increased to 72.4% after integrating EnKF. Recall improved from 52.8% to 76.9%, and the F1 Score rose from 0.62 to 0.83. Moreover, the IoU values ranged from 0.93 to 0.96, indicating high precision in object bounding box predictions. These findings confirm that integrating YOLOv5 with EnKF effectively enhances object detection performance, particularly in complex scenarios. The study highlights the potential benefits of combining deep learning algorithms like YOLOv5 with statistical methods such as EnKF to improve the robustness and effectiveness of object detection systems.