Open AccessDeep learning based target detection algorithm for motion capture applications
Author(s) -
Haitao Wang,
Xin Tong,
Fengyun Lu
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1682/1/012032
Subject(s) - computer science , motion capture , artificial intelligence , mobile device , deep learning , motion (physics) , convolutional neural network , computer vision , automatic identification and data capture , orientation (vector space) , field (mathematics) , algorithm , geometry , mathematics , speech recognition , operating system , pure mathematics
Motion capture technology is the use of external devices to perform data recording and posture reproduction of the displacement of human structures. Deep learning algorithms are playing an increasingly important role in motion capture technology as the technology involves data that can be directly understood and processed by computers in terms of dimensional measurements, positioning of objects in physical space, and orientation determination. This paper presents an application of a convolutional neural network system, YOLO-V4, in the field of motion capture. YOLO-V4 system weight files are small and do not require high hardware requirements. It can also be implemented in PyTorch so that it can be deployed on mobile devices, enabling edge devices to run these models as well, relieving the space constraint of immovable signal capture devices and providing the advantages of high accuracy and high detection rate.