• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>International Symposium on Systems and Human Science; 20031119-20; Osaka(JP) >Tracking People and Action Recognition from Omnidirectional Images
【24h】

Tracking People and Action Recognition from Omnidirectional Images

机译：从全向图像跟踪人员和动作识别

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

In this chapter, we propose methods for tracking people and recognizing their actions through indoor scenes captured with an omnidirectional image sensor. They can be used to detect and track people to extract their trajectories of movement and their actions are then recognized by using extracted trajectories. Recently, stochastic algorithms have frequently been used for action recognition because they require non-linear and non-Gaussian models of action. Action models prepared from trajectories, however, include movements that are almost the same, so they are redundant. We have, therefore, assumed that human actions can be classified into action primitives, which are modeled by transitions of discrete states considered as action primitives. The methods we propose combine continuous state models and discrete state models by stochastic sampling generated from state transition probabilities.
机译：在本章中，我们提出了通过全向图像传感器捕获的室内场景跟踪人员并识别人员行为的方法。它们可用于检测和跟踪人们以提取他们的运动轨迹，然后通过使用提取的轨迹来识别他们的动作。近年来，随机算法经常被用于动作识别，因为它们需要非线性和非高斯的动作模型。但是，从轨迹准备的动作模型包含几乎相同的动作，因此是多余的。因此，我们假设人类行为可以分类为动作原语，这些动作原语是通过将离散状态转换为动作原语来建模的。我们提出的方法通过根据状态转移概率生成的随机采样将连续状态模型和离散状态模型结合在一起。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

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