An Adaptive Appearance Template Updating Method for Online Multi-Object Tracking
Electronics · 17.07.2026 · 15 (14) · 3151
DOI: 10.3390/electronics15143151 ↗Kurz gefasst
Fachartikel in der Zeitschrift Electronics, online erschienen am 17.07.2026. Verfasst von Jiayuan Wan und 1 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Online appearance-template updating affects the stability of subsequent data association in multi-object tracking. In StrongSORT, the appearance template is updated with a fixed-coefficient exponential moving average (EMA). Under occlusion, false detections, or feature noise, this strategy can write abnormal observations into the historical template and consequently increase identity switches. Without changing the matching architecture or track-management rules of StrongSORT, this study modifies only the appearance-template updating stage and proposes an adaptive appearance-template updating method. The proposed method estimates observation reliability from appearance consistency, motion consistency, track maturity, and detection confidence. The estimated reliability is then used to dynamically adjust the retention coefficient of the historical appearance template, allowing reliable observations to be incorporated more rapidly. Furthermore, a soft-freeze weak-updating mechanism is introduced for potentially unreliable observations, in which a higher historical-template retention coefficient is used to suppress template contamination from abnormal features. Ablation experiments on the MOT17 ablation split show that, while multiple-object tracking accuracy (MOTA) remains nearly stable, the proposed method reduces the number of identity switches of StrongSORT from 296 to 206, corresponding to a reduction of approximately 30.4%. The evidence therefore supports improved suppression of discrete identity-switch events under the evaluated StrongSORT setting, rather than a uniform improvement across all tracking metrics.
Zitieren
APA
Wan, J., & Ni, H. (2026). An Adaptive Appearance Template Updating Method for Online Multi-Object Tracking. Electronics, 15(14), 3151. https://doi.org/10.3390/electronics15143151
Vancouver
Wan J, Ni H. An Adaptive Appearance Template Updating Method for Online Multi-Object Tracking. Electronics. 2026;15(14):3151. doi:10.3390/electronics15143151
BibTeX
@article{Wan2026,
author = {Wan, Jiayuan and Ni, Hongjie},
title = {An Adaptive Appearance Template Updating Method for Online Multi-Object Tracking},
journal = {Electronics},
year = {2026},
volume = {15},
number = {14},
pages = {3151},
doi = {10.3390/electronics15143151}
}
EUVIMED