Visual Hull Reconstruction For Automated Primate Behavior Observation
Nastaran Ghadar, Xikang Zhang, Kang Li, Guillaume Thibault, Alireza Bayesteh, Izhak Shafran, Deniz Erdogmus, Kris Coleman, Kathleen A. Grant

Abstract:
The study of social animal interactions is used as means for understanding animal behavior and biology. In this work, we describe a computerized method that utilizes 3D visual hull reconstruction to identify and localize rhesus macaques in their social groups. There are three major steps in this study. First, we collect experimental data from four synchronized cameras at different locations and angels in a cage containing five rhesus macaques. Second, by using computer vision algorithms, we detect and identify animals using 2D observations that were provided from the previous step. This provides essential quantitative data for animal behavior research. Finally, by applying visual hull reconstruction algorithm, we automatically build a 3D model for each rhesus macaques on every frame. The results of this work can be used for tracking these animals in their cage, and furthermore it can be used for activity recognition of social interactions of rhesus macaques. The method we developed in this paper, shows promising results that are accurate, yet runs in a timely manner ; this makes this algorithm suitable for large datasets and we can use it for future high-level recognition tasks.