Automated Tail Position Tracking with Millimeter Accuracy using Depth Sensing and Mask R-CNN

Abstract

Despite a strong belief and some early evidence of dogs' use of their tail in communication, very little is actually known with confidence about what is being communicated and how. In part this lack of knowledge is likely a result of immature tools available to researchers desiring to study dogs' tail communications. To address this tool gap, we have developed an image processing pipeline using a depth camera, a deep neural network for image segmentation, and pointclouds that enables autonomous extraction of tail position and movement analytics with high fidelity. We present the pipeline and segmentation approach, as well as present some data from a feasibility study. The data are encouraging and indicate promise for this technique, albeit they are not yet conclusive.