Researchers develop AI method to remove humans from egocentric walking tour videos for cleaner environment modeling

Novel generative algorithm automatically removes humans and shadows from first-person perspective walking tour videos to improve environment visual models

Team created semi-synthetic training dataset by pairing real environment backgrounds with composite clips of walking humans and simulated shadows

Uses randomly sourced foreground and background components from authentic egocentric walking tour videos worldwide to maintain visual diversity

Addresses key limitation of walking tour videos for environment modeling: excessive human presence from crowds and eye-level camera perspectives