Robust tracking and posture description for laboratory rodents using active shape models

Abstract

We are in the process of developing an automated image analysis system, which uses deformable models of shape, learned from image examples, to interpret video images of rodents. Active shape models provide a compact description of the shape of the animal in a way that enables the postures that differentiate various behaviors to be distinguished. They also model the image profile across the shape boundary. We show how these features allow automatic, robust segmentation of the explicit object of interest. Rather than just detecting movement or changes from background in the image, the system can focus on objects that are of the correct shape and appearance. The modeling of the image profiles also allows the system to distinguish between the actual animal and image artifacts. We show how these techniques are being extended to extract postural information, which can then be integrated with positional data to produce a model of behavior.