Multi-Class Multi-Object Tracking in Aerial Images Using Uncertainty Estimation

다국어 초록 (Multilingual Abstract)

Multi-object tracking (MOT) is a vital component in understanding the surrounding environ -ments. Previous research has demonstrated that MOT can successfully detect and track surroundingobjects. Nonetheless, inaccurate classification of the tracking objects remains a challenge that needs to besolved. When an object approaching from a distance is recognized, not only detection and tracking butalso classification to determine the level of risk must be performed. However, considering the erroneousclassification results obtained from the detection as the track class can lead to performance degradationproblems. In this paper, we discuss the limitations of classification in tracking under the classificationuncertainty of the detector. To address this problem, a class update module is proposed, which leveragesthe class uncertainty estimation of the detector to mitigate the classification error of the tracker. Weevaluated our approach on the VisDrone-MOT2021 dataset, which includes multi-class and uncertain far-distance object tracking. We show that our method has low certainty at a distant object, and quicklyclassifies the class as the object approaches and the level of certainty increases. In this manner, our methodoutperforms previous approaches across different detectors. In particular, the You Only Look Once(YOLO)v8 detector shows a notable enhancement of 4.33 multi-object tracking accuracy (MOTA) incomparison to the previous state-of-the-art method. This intuitive insight improves MOT to trackapproaching objects from a distance and quickly classify them.