Title:
RECOGNITION DEVICE, LEARNING DEVICE, RECOGNITION SYSTEM, RECOGNITION METHOD, RECOGNITION PROGRAM, LEARNING METHOD, AND LEARNING PROGRAM
Document Type and Number:
Japanese Patent JP2021114223
Kind Code:
A
Abstract:
To calculate an integral feature amount with high approximate accuracy at high speed, and recognize a recognition object at high speed with high accuracy.SOLUTION: Feature amount extraction means 50 extracts feature amounts for each divided region of an input image using convoluted neural network for extracting the feature amounts for each of the divided regions with an input image as input. Weight calculating means 51 calculates a weight according to multiplicity for each of the divided regions overlapping an inspection region. Integral means 52 determines an integral feature amount obtained by integrating a feature amount extracted from each of the divided regions overlapping the inspection region, using the calculated weight. Recognition means 56 recognizes whether or not the inspection region indicates the recognition object using the integral feature amount.SELECTED DRAWING: Figure 6
Inventors:
SANO YUSUKE
ITOGA TAKESHI
ITOGA TAKESHI
Application Number:
JP2020007381A
Publication Date:
August 05, 2021
Filing Date:
January 21, 2020
Export Citation:
Assignee:
SECOM CO LTD
International Classes:
G06N3/04; G06T7/00
Domestic Patent References:
| JP2019016298A | 2019-01-31 |
Foreign References:
| WO2014030399A1 | 2014-02-27 |
Other References:
"映像からのオブジェクト識別技術", ＮＨＫ技研Ｒ＆Ｄ ＮＯ．１４２, JPN6023045821, 15 November 2013 (2013-11-15), ISSN: 0005190260
Attorney, Agent or Firm:
Patent Service Corporation Taiyo International Patent Office