Title:
METHOD AND SYSTEM FOR EXTRACTING, STORING AND RETRIEVING MASS VIDEO FEATURES
Document Type and Number:
WIPO Patent Application WO/2020/147857
Kind Code:
A1
Abstract:
Disclosed are a method and a system for extracting, storing and retrieving mass video features, wherein feature extraction is performed by training a deep convolutional neural network, the storage method comprises the steps of video slicing, interval sampling, landmark recognition, deep convolutional neural network processing to obtain hash codes and features, selection of key frames and the like, the retrieval method comprises landmark information and picture retrieval, when a picture is input, firstly extracting the hash codes and features through the deep convolutional neural network, quickly retrieving the hash codes in a hash index database, then accurately matching the features in a video feature database, and obtaining video information in a video information database after obtaining the index of a similar landmark; and the present application has advantages of better accuracy and high precision.
Inventors:
LI CHUANPENG (CN)
GU YINZHENG (CN)
XIE JINBIN (CN)
GU YINZHENG (CN)
XIE JINBIN (CN)
Application Number:
PCT/CN2020/072969
Publication Date:
July 23, 2020
Filing Date:
January 19, 2020
Export Citation:
Assignee:
SHANGHAI JI LIAN WEB TECH CO LTD (CN)
International Classes:
G06F16/73; G06K9/00
Domestic Patent References:
| WO2018086513A1 | 2018-05-17 |
Foreign References:
| CN109815364A | 2019-05-28 | |||
| CN108280233A | 2018-07-13 | |||
| CN107943938A | 2018-04-20 | |||
| CN109086690A | 2018-12-25 |
Attorney, Agent or Firm:
BEYOND ATTORNEYS AT LAW (CN)
Download PDF: