In-vehicle illumination-invariant change detection based on intrinsic images and di erences of Gaussians

摘要

Image change detection methods allow to automatically detect changes on outdoor routes with respect to aprevious time point and to improve the identification of potential threats for vehicles (e.g., improvised explosivedevices - IEDs). The change detection task is challenging, mainly due to the arbitrary type of changes and dueto the global and local illumination differences. We have developed an illumination-invariant change detectionapproach based on intrinsic images and differences of Gaussians. The intrinsic images are obtained by the logchromaticityspace of color images and in conjunction with reectance images obtained by homomorphicfiltering,illumination-invariant differences images are determined. We employ differences of Gaussians to detect blobs ofmultiple sizes in the differences images, which describe detected changes. To reduce false positives we proposeto weight the scores of the detected blobs based on normalized cross-correlation, and to enforce temporallyconsistent changes we employ a temporal filtering scheme. The approach has been successfully applied to imagesequences acquired on outdoor routes at different time points, demonstrating the consistent detection of changesover time, and the invariance to illumination differences, including shadows.