Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
MECHANICAL ARM MOTION CAPTURE METHOD, MEDIUM, ELECTRONIC DEVICE, AND SYSTEM
Document Type and Number:
WIPO Patent Application WO/2023/168849
Kind Code:
A1
Abstract:
A mechanical arm motion capture method, comprising: S1, fixing, as a source domain on a mechanical arm, data acquired by a visual sensor, fixing, as a target domain corresponding to a human arm, data acquired by an inertial sensor, and establishing a state space expression of the system; S2, setting optimal unknown state observation joint distribution by using a total probability theory and taking source domain observation prediction distribution as a condition, decomposing a condition joint observation distribution model, and solving optimal distribution by using KL divergence; and S3, transferring, by combining a Kalman filter with the total probability theory, source domain knowledge measured by the visual sensor into the target domain measured by the inertial sensor, performing Kalman filtering-based data fusion, and predicting the state of the system at a next moment so as to realize motion capture of the mechanical arm. The defects of visual sensing and inertial sensing are comprehensively considered, and by means of a thought of transfer learning, the visual sensing is used as the source domain so as to improve the target domain of the inertial sensing, thereby improving the prediction accuracy.

Inventors:
ZHAO SHUNYI (CN)
WANG WEI (CN)
LUAN XIAOLI (CN)
LIU FEI (CN)
Application Number:
PCT/CN2022/099469
Publication Date:
September 14, 2023
Filing Date:
June 17, 2022
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
UNIV JIANGNAN (CN)
International Classes:
B25J9/16
Foreign References:
CN114454174A2022-05-10
CN103112007A2013-05-22
CN106182003A2016-12-07
CN111890373A2020-11-06
US11254002B12022-02-22
Attorney, Agent or Firm:
CENTRAL SOUTH WELL INTELLECTUAL PROPERTY OFFICE (CN)
Download PDF: