Understanding Identity Continuity in Thermal Video through Scene-Level Consistency
Researchers have developed a method to improve identity continuity in thermal video pedestrian tracking. Their approach focuses on lightweight post-processing techniques rather than complex re-identification models. By adding a modular backend for remapping short gaps and relinking tracklets using spatial, temporal, and motion cues, they enhanced the IDF1 score from 82.25 to 84.93 on the PBVS Thermal Pedestrian MOT benchmark. AI
IMPACT Enhances tracking accuracy in thermal imagery, potentially improving surveillance and autonomous system perception in low-visibility conditions.