Visual tracking of small animals based on real-time Level Set Method with fast infra-red thermographic imaging

In this paper, we present a long-term monitoring of moving and high deformable objects with fast thermographic infrared imaging, in a process of developing an automated image analysis of laboratory animals' postures and activities. For this several days tracking, we tackle the object deformations with a model-based articulated body parts scheme and the two-stage tracking and segmentation technique is performed through a Kalman filtering associated to a level-set method. The latter is the main contribution of the paper (with also the contours classification of each body parts) since the proposed segmentation is robust to noise and heat places and can easily be parallelized and executed in real-time. Each body part is modelled with a set of geometrical descriptors with mutual interactions. The advantage of such method is not only to separate the dynamics of the whole object motion from those of body parts deformations but it also allows to extract apparent postural information to further produce a model of animal behaviour.