Biologically-inspired robust motion segmentation using mutual information

6 This paper presents a neuroscience inspired information theoretic approach 7 to motion segmentation. Robust motion segmentation represents a funda8 mental first stage in many surveillance tasks. As an alternative to widely 9 adopted individual segmentation approaches, which are challenged in differ10 ent ways by imagery exhibiting a wide range of environmental variation and 11 irrelevant motion, this paper presents a new biologically-inspired approach 12 which computes the multivariate mutual information between multiple com13 plementary motion segmentation outputs. Performance evaluation across a 14 range of datasets and against competing segmentation methods demonstrates 15 robust performance. 16