Facial Landmark Localization Optimization for Live Streaming Scenarios
IET Image Processing · 2026-01 · 20 (1)
DOI: 10.1049/ipr2.70421 ↗Kurz gefasst
Fachartikel in der Zeitschrift IET Image Processing, online erschienen am 29.07.2026. Verfasst von Di Wang und 2 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Practical pain points of live streaming object detection and tracking algorithms include interference from complex poses and viewing angles, normalized various occlusion problems, truncation at frame edges, fluctuations of illumination and image quality, poor adaptation to offline template videos, and conflicts between lightweight deployment and scene adaptation. The research value in industrial deployment includes enabling real‐time virtual human interaction, supporting industrial applications of beauty retouching and filter effects, reducing development and computational costs, meeting privacy and commercial data compliance requirements and providing basic support for live‐streaming AI applications. Practical impacts caused by technical defects in live streaming scenarios include virtual human interaction, beauty and special effects, system stability, and engineering deployment cost. To address the problems of low accuracy and poor robustness in facial landmark localization caused by complex conditions, such as partial body appearance and large pose variations in live streaming scenarios, this paper proposes a joint optimization scheme for face detection and facial landmark localization tailored for live streaming. Firstly, a spatial constraint relationship between the human body and face is constructed, and prior information from human body keypoints is utilized to assist to improve the accuracy of face detection and localization. Secondly, a dedicated dataset containing 1500 live streaming video clips is annotated and constructed to solve the problem of the insufficient generalization ability of general datasets. Finally, a scene‐adaptive data augmentation strategy is designed. Experimental results on the self‐built live streaming dataset demonstrate that the comprehensive performance of the proposed algorithm outperforms mainstream detectors such as RetinaFace and YOLO‐v11. The proposed scheme can be effectively applied to virtual human live streaming, live streaming beauty effects, virtual prop interaction, and other scenarios, showing high practical value.
Zitieren
APA
Wang, D., Shi, B., & Zhu, W. (2026). Facial Landmark Localization Optimization for Live Streaming Scenarios. IET Image Processing, 20(1). https://doi.org/10.1049/ipr2.70421
Vancouver
Wang D, Shi B, Zhu W. Facial Landmark Localization Optimization for Live Streaming Scenarios. IET Image Processing. 2026;20(1). doi:10.1049/ipr2.70421
BibTeX
@article{Wang2026,
author = {Wang, Di and Shi, Boqiang and Zhu, Wenying},
title = {Facial Landmark Localization Optimization for Live Streaming Scenarios},
journal = {IET Image Processing},
year = {2026},
volume = {20},
number = {1},
doi = {10.1049/ipr2.70421}
}
EUVIMED