Title:
GESTURE RECOGNITION METHOD, DEVICE, ELECTRONIC APPARATUS, AND STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2020/140798
Kind Code:
A1
Abstract:
A gesture recognition method, a device, an electronic apparatus, and a storage medium. The gesture recognition method comprises: acquiring a hand image (S101); adopting a feature mapping model, and extracting a first standard feature of a hand, the feature mapping model being obtained from pre-training with respect to second standard features of hands in synthesized image samples and real image samples (S102); processing the first standard feature, and obtaining three-dimensional coordinates of multiple key points of the hand (S103); and determining a hand gesture according to the three-dimensional coordinates of each of the key points (S104). In the method, common hand features are adopted from synthesized images and real images to perform model training, thereby eliminating differences between synthesized images and real images in terms of standard features, and accordingly improving the accuracy of hand gesture recognition in weak supervision scenarios. The method does not require synthesized image samples and real image samples to have an entirely consistent gesture, thereby relaxing requirements on training data sets. The method is further applicable to gesture recognition using color images and thus has a wider range of applications.
Inventors:
LIU YUFENG (CN)
ZHENG WEN (CN)
ZHENG WEN (CN)
Application Number:
PCT/CN2019/128029
Publication Date:
July 09, 2020
Filing Date:
December 24, 2019
Export Citation:
Assignee:
BEIJING DAJIA INTERNET INFORMATION TECH CO LTD (CN)
International Classes:
G06V10/774
Foreign References:
| CN109858524A | 2019-06-07 | |||
| CN108062526A | 2018-05-22 | |||
| CN107103613A | 2017-08-29 | |||
| CN104598915A | 2015-05-06 | |||
| CN108229318A | 2018-06-29 | |||
| US20150378444A1 | 2015-12-31 |
Attorney, Agent or Firm:
TDIP & PARTNERS (CN)
Download PDF: