首页> 外国专利> ACTION RECOGNITION APPARATUS, ACTION LEARNING APPARATUS, ACTION RECOGNITION PROGRAM, AND ACTION LEARNING PROGRAM

ACTION RECOGNITION APPARATUS, ACTION LEARNING APPARATUS, ACTION RECOGNITION PROGRAM, AND ACTION LEARNING PROGRAM

机译:动作识别装置,动作学习装置,动作识别程序和动作学习程序

摘要

According to the present invention, during action recognition learning and action recognition, a data amount used for action recognition is determined through stable and appropriate consideration of actions performed before the action at a current recognition time instead of consideration of the time length (number of frames), to improve the accuracy and the efficiency of the action recognition. This action recognition apparatus has a recognition unit (12) that recognizes actions of a target such as a person, on the basis of time-sequential data of the feature amount of actions of the target, extracted from data (moving image data) in which the actions are time-sequentially recorded. The recognition unit recognizes an action at each time by time-sequentially tracking the time-sequential data of the feature amount, sets a series of single actions that are not discriminated from one another as one action, and after the recognition of a prescribed number of actions that are aligned in time sequence, recognizes an action at the current recognition time on the basis of the time-sequential data of the feature amount equivalent to a plurality (three in FIG. 8) of actions aligned in time sequence from the time before the prescribed number of actions to the current recognition time (802, 804).
机译:根据本发明,在动作识别学习和动作识别期间,通过稳定且适当地考虑在当前识别时间在动作之前执行的动作来确定用于动作识别的数据量,而不是考虑时间长度(帧数) ),以提高动作识别的准确性和效率。该动作识别装置具有识别单元(12),该识别单元(12)基于从数据(运动图像数据)中提取的目标的动作的特征量的时间序列数据来识别目标,例如人。这些动作是按时间顺序记录的。识别单元通过按时间顺序跟踪特征量的时间序列数据来每次识别一个动作,将一系列彼此不区分的单个动作设置为一个动作,并在识别出规定数量的按时间顺序排列的动作,基于特征量的时间序列数据,在当前识别时间识别动作,该特征量的时间序列数据与从前一个时间开始按时间顺序排列的多个动作(图8中的三个)相等到当前识别时间的规定次数的动作(802、804)。

著录项

  • 公开/公告号WO2017150211A1

    专利类型

  • 公开/公告日2017-09-08

    原文格式PDF

  • 申请/专利权人 KONICA MINOLTA INC.;

    申请/专利号WO2017JP05850

  • 申请日2017-02-17

  • 分类号G06T7/20;G06T7;

  • 国家 WO

  • 入库时间 2022-08-21 13:29:57

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号