Student: Christina Smith
Project Mentors: Dr. Christopher Buneo – SBHSE
Dr. Claire Honeycutt – SBHSE
Dr. Stephen Helms Tillery – SBHSE
YouTube Link: View the video link below before joining the zoom meeting
Zoom Link: https://asu.zoom.us/j/5381252533
Zoom meeting time: 9am - 11am
Abstract
Deep learning is a sub-category of a machine learning method that is rooted in artificial neural networks with representation learning. The reason for the creation of deep learning is due to the limitation of conventional machine-learning techniques in their ability to process natural data in their raw form. An approach that has researchers and clinicians interested is the application of deep learning in biological research to integrate vast datasets and arbitrarily complex relationships. In order to apply this to biological research involving motion capture of human and animal subjects, a software toolkit called DeepLabCut was created. It relies on the use of deep learning to identify movements and patterns without the need for active or passive markers. Specifically, this software is advantageous to use when capturing animal movement to model their behavior. Therefore, the purpose of this project was to investigate this toolkit and its applicability to be integrated into a laboratory setting that uses animal and human models in its research. The project was divided into two sections: 1) identification of optimal parameters and 2) capturing and analyses of original data. First, three parameters – displayed iterations, saved iterations, and maximum iterations – were investigated in terms of how well the network was trained and the time investment to train it. After these parameters were defined and set, we captured original data of repetitive human arm motions to determine how accurately the software toolkit can track them. In addition, we determined what data can be extracted from a capture session by running analyses on the data using MATLAB code. From this project, we hope to provide more knowledge about this cost-efficient motion capture system and its applicability to laboratory research.