Identifying semi-Invariant Features on Mouse Contours

This paper addresses the problem of reliably fitting an orientated model to video data of laboratory mice assays by specifically locating semi-invariant points on an extracted outline. In the case of mice, the rapid changes in direction and shape often lead to failure when using explicit models. Here we employ a standard background subtraction algorithm in order to derive contour information from a well defined top-down view of the assay. Using this contour, we compare three different approaches at locating head, tail-tip and tail-base features that allow us to constrain orientation. We validate each approach against an annotated gold-standard data-set, and conclude that a composite method delivers the best results. This ultimately has benefits for analysing higher-level behaviour where it is crucial to retain orientation.