Application of a depth image analysis algorithm to automatic social behaviour estimation of laboratory animals

The problem of animal social behavior estimation has two components: automatically tracking a number of identical objects with a large number of collisions and calculating indexes for behavior description. In this paper a tracking algorithm has been proposed for a sequence of depth images. This algorithm combines detection and tracking methods. The proposed algorithm is highly reliable: in the worst case it yields one error per 193 collisions. Several useful automatically estimated indexes for social behavior were introduced in this paper.