Change Detection in Presence of Registration Error Using LTP Histograms

摘要

This paper deals with detection of significant changes between two images of the same indoor scene acquired from different positions. The nonstatic camera position leads to the need of a registration technique which may be inaccurate. Moreover in case of non-planar scenes accurate registration is not possible. To deal with such inaccuracy a method based on Local Ternary Patterns (LTP) is proposed. Tests performed on real data show that the proposed method is able to detect the changed region in the scene while neglecting the registration error.