본문 바로가기
  • Home

Comparison of Tracking-by-detection Algorithm on Aerial Battle Scenario based On-Device

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2025, 30(12), pp.287~296
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : September 4, 2025
  • Accepted : December 11, 2025
  • Published : December 31, 2025

Du-Hwan Hur 1 Dae-Hyeon Park 1 Deok-Woong Kim 1 Junhyeong Bak 2 Ingu Park 2 Ki-Nam Lee 2 Seung-Hwan Bae 1

1인하대학교
2엘아이지넥스원(주)

Accredited

ABSTRACT

In this paper, we present a guideline for multi-object tracking in aerial battle scenarios, where tracking is required from low-resolution infrared imagery and low-quality detection results. Albeit many MOT algorithms are compared on high-performance resource (e.g., high-performed devices, high-resolution inputs and powerful detectors), the analysis on the resource-constrained scenario is limited. To resolve this gap, we analyze how such resource-constrained conditions affect tracking performance and computational cost. We demonstrate that powerful detections leads to improved Average IoU and reduced false positives (FP) in the tracker. We also compare recent tracking-by-detection under both high and low resolution conditions. Based on these results, we provide tracking algorithm that clarify whether to reduce false positives (FP) or false negatives (FN). As a result, we provide the guideline in aerial battle scenarios, demonstrating that tracking accuracy depends on detection confidence and that the association algorithms focus on regulating both FP and FN.

Citation status

* References for papers published after 2024 are currently being built.