Markerless 3D Pose Estimation System for Mouse Musculoskeletal Model Using DeepLabCut and Multiple RGB-D Cameras
Yoshito TSURUDA, Shingo Akita, Kotomi Yamanaka, Masataka Yamamoto, Yoshitake Sano, Teiichi Furuichi, Hiroshi Takemura · 2023 IEEE/SICE International Symposium on System Integration (SII) · 2023
Animal experiments play an important role in basic research as well as in applied research such as new drug development. In particular, the mouse is one of the most valuable laboratory animals in medical research because of its many advantages, such as its small size and ease of breeding. However, evaluations of behavioral animal experiment analysis are often conducted visually by observers, and there are problems with subjective evaluation and human error. An objective and automatic mouse analysis system is needed to solve these problems. Although several analysis systems have been developed in recent years, these systems have problems such as rough behavioral classification and attachment of markers that may affect mouse behavior. In this study, we proposed a detailed behavioral analysis system for a mouse without markers using six RGB-D cameras and video tracking based on deep learning. As a result, estimation with an average error of 5 to 10 mm at keypoints was achieved with this system.