基于深度强化学习的计步方法

被引：4

作者：

彭琛

韩立新

机构：

[1] 河海大学计算机与信息学院

来源：

计算机与现代化 | 2019年 / 01期

关键词：

计步器; 深度强化学习; 均值穿越波峰波谷检测法;

D O I：

暂无

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

针对计步软件使用中用户行为不定,容易产生各种噪声以及传统算法中参数不能持续优化的问题,本文提出基于深度强化学习的计步方法。将噪声判别及步数统计作为智能体的动作,在步数统计中改进波峰检测法,提出均值穿越波峰波谷检测法。利用循环神经网络保存内部状态,将用户对计步器计步好坏的反馈作为奖励信号,指导参数持续优化。实验结果表明,该方法在采集设备放置于不同位置并且有噪声时,噪声识别率为0.9151,计步误差率为0.0623,有较高的精度以及较强的抗干扰能力。

引用

页码：63 / 68

页数：6

共 24 条

[1]

Deep learning in neural networks: An overview[J] . Jürgen Schmidhuber. &nbspNeural Networks . 2014

[2]

Learning Hand-Eye Coordination for Robotic Grasping with Large-Scale Data Collection. LEVINE S,PASTOR P,KRIZHEVSKY A,et al. International Journal of Robotics Research . 2016

[3]

End-to-end training of deep visuomotor policies. Levine S,Finn C,Darrell T,et al. Journal of Machine Learning Research . 2015

[4]

Autonomous footstep counting and traveled distance calculation by mobile devices incorporating camera and accelerometer data. LU Y T,VELIPASALAR S. IEEE Sensors Journal . 2017

[5]

Walk detection and step counting on unconstrained smartphones. BRAJDIC A,HARLE R. The International Joint Conference on Pervasive and ubiquitous computing . 2013

[6]

Target-driven visual navigation in indoor scenes using deep rein forcement learning. Zhu Yuke,MOTTAGHI R,KOLVE E,et al. IEEE International Conference on Robotics&Automation,ICRA . 2017

[7]

ERSP:An energy-efficient real-time smartphone pedometer. OSHIN T O,POSLAD S. IEEE International Conference on Systems,Man,and Cybernetics (SMC) . 2013

[8]

Self-Adaptive Step Counting on Smartphones under Unrestricted Stepping Modes. Tang Z,Guo Y,Chen X. Computer Software and Applications Conference . 2016

[9]

Step count and classification using sensor information fusion. ANACLETO R,FIGUEIREDO L,ALMEIDA A,et al. Ambient Intelligence-Software and Applications . 2015

[10]

A step counting algorithm for smartphnone users:Design and implementation. PAN M S,LIN H W. Sensors Journal . 2015

← 1 2 3 →