본문 바로가기

Low-Light Face Recognition for Mobile Robots

문서 유형
연구출판물 · 학술대회 발표논문
등재
Scopus
피인용
Scopus 0
FWCI (Scopus)
0

저자

  1. 백창우제1저자교신저자Pusan National University, Dept. of Electronics Engineering, Busan, Korea
  2. Song, Jou WonLG Electronics, Souel, Korea
  3. 공경보Pusan National University, Dept. of Electronics Engineering, Busan, Korea

초록

Face recognition models achieve high accuracy in controlled environments but suffer significant performance degradation under real-world conditions such as low illumination and motion blur - common in robotic applications. To address this challenge, we explore two complementary strategies: (1) low-light enhancement (LLE) via image preprocessing, and (2) data-centric augmentation using a controllable face synthesis approach that simulates diverse low-quality conditions while preserving facial identity. We conduct extensive experiments on a mobile robot platform under various lighting scenarios - normal, low-light, and backlight. The results show that while LLE provides limited and inconsistent benefits, the controllable synthesis-based augmentation consistently improves recognition performance across all scenarios without adding inference-time overhead. These findings suggest that training-time augmentation using identity-preserving, low-quality face synthesis offers a more robust and practical solution than preprocessing for real-world robotic face recognition.

키워드

Face recognitionLow-light image enhancementRobotic visionStyle transfer

식별자

Scopus EID
2-s2.0-105016361998