Whisker Analysis Framework for Unrestricted Mice with Neural Networks

Zhijie Tan,Shengwei Meng,Yujia Tan,Yiqun Wu,Yi Li,Tong Mo,Weiping Li

DOI: https://doi.org/10.1007/978-3-031-44192-9_40

2023-01-01

Abstract:Whiskers of rats and mice form an important tactile system. Non-invasive analysis of the movement pattern of whiskers can inspire the development of research including the tactile perception of rodents. However, the whiskers of mice are very slender and easily confused with body hair or video background. At the same time, the vibration frequency of whiskers is very high, and the vibration process may produce artifacts or motion blurs. In addition, the movement pattern analysis of whiskers needs to exclude the ego-motion of the head for unrestricted mice. In this paper, we propose WAFUM (Whisker Analysis Framework for Unrestricted Mice) to provide a visual analysis model for the movement pattern of whiskers in a non-invasive way with neural networks. In the first step, this model detects the region where the whiskers are located, makes semantic segmentation of the whiskers, and generates the pose of the whisker group by clustering. The second step is to recover the head pose through key point detection and tracking. Finally, the model combines the results of the former two steps to recover the ego-motion mode of whiskers. The experimental part demonstrates that with the help of neural networks, whisker tracking for the unrestricted mice can be completed using a consumer-grade smartphone in a non-invasive way while all of the other fragile traditional algorithms fail.