Interior defect detection through vehicle glazing using multi-view motion analysis and asymmetric illumination

Abstract

A method and system for detecting interior defects in vehicles through glazing without requiring door opening or interior sensor installation. The system captures multi-view images of vehicle window panes from spatially separated viewpoints during relative motion between vehicle and imaging device. Motion field data representing pixel displacement is computed and used to classify pixels into interior-origin pixels exhibiting apparent depth behind the window pane plane and reflection-origin pixels exhibiting apparent depth at the glass surface. An interior-enhanced image is generated through motion refocusing that sharpens interior textures while blurring glass-plane reflections. Optional asymmetric illumination and polarization provide additional reflection suppression. Transmittance compensation accounts for glazing tint and spectral characteristics. Defect detection algorithms identify interior defects including seat damage, missing components, and structural defects, with privacy-aware redaction of occupants integrated into the processing pipeline.