• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>International Conference on Cloud Computing and Security >HL-HAR: Hierarchical Learning Based Human Activity Recognition in Wearable Computing
【24h】

HL-HAR: Hierarchical Learning Based Human Activity Recognition in Wearable Computing

机译：HL-HAR：可穿戴计算中基于分层学习的人类活动识别

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

In recent years there have been many successes of recognizing the human activity using the data collected from the wearable sensors. Besides, many of these applications use the data from the smartphone. But it is also a challenge in practice for two reasons. Most method can achieve a high precision in the cost of increasing memory consumption, or asking for complicated data source. In this paper, (1) Utilizing Plus-L Minus-R selection to single out the optimal combination from the feature vector extracted; (2) Introducing a fast classification method named H-ELM to resolve the problem of the highly memory consumption in the process of calculation. The main benefit of this factor is to reduce memory usage and increase recognition accuracy with a brief feature vector so that a wearable device can identify activities all by itself. And the wearable device can recognize the sample activities even if keeping away from cellphone. Our results show that this method leads to that we can recognize object activities with the overall accuracy of 93.7% in a very short period of time on the dataset of Human Activity Recognition Using Smartphones Dataset. The selected 25-dimension feature vector nearly contains all the information and after many times of test, it can achieve very high percentage of accuracy. Moreover, the method enables the learning velocity to outperform the state-of-the-art on the Human Activity Recognition domain.
机译：近年来，使用从可穿戴式传感器收集的数据来识别人类活动已经取得了许多成功。此外，许多这些应用程序都使用智能手机中的数据。但这在实践中也是一个挑战，原因有两个。大多数方法可以在增加内存消耗或要求复杂的数据源的成本中实现高精度。本文的研究：（1）利用Plus-L Minus-R选择从提取的特征向量中选出最优组合; （2）引入一种称为H-ELM的快速分类方法，以解决计算过程中内存消耗高的问题。此因素的主要好处是可以通过简短的特征向量来减少内存使用量并提高识别精度，从而使可穿戴设备可以自行识别所有活动。而且即使远离手机，可穿戴设备也可以识别样本活动。我们的结果表明，该方法导致我们可以在很短的时间内在使用智能手机数据集进行人类活动识别的数据集上以93.7％的总体准确度识别对象活动。所选的25维特征向量几乎包含所有信息，并且经过多次测试，可以达到很高的准确率。此外，该方法使学习速度能够胜过人类活动识别领域的最新技术。

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