Non-rigid Tracking of Musk Shrews in Video for Detection of Emetic Episodes

Abstract

Animal study plays a key role in understanding the mechanisms of emesis induced by many diseases and drug treatments. The traditional measurement of emesis usually requires manual observation which is labor intensive, difficult to standardize across coders, and often lacks spatial and temporal precision. This paper proposes a vision-based system for non-rigid tracking of musk shrews in videos and automatic characterization of eme-sis response when a chemotherapy agent is applied. Firstly, we learn a 80-dimensional non-rigid shape model (defined by 40 two-dimensional landmarks) to characterize the non-rigid shape variation of a musk shrews. After the body contour of the musk shrew is initialized in the first frame, we propose an algorithm for non-rigid tracking of the 40 landmarks in a video sequence. In each frame the contour of the musk shrew is estimated using background subtraction and a major challenge is to solve for the correspondence between the body contour and the shape model. Once the tracking is done, we used the shape changes to characterize the retch response during the emetic episodes. Finally, an emetic episode is detected when there're consecutive retches.