【作者】 杨森；

【导师】 李明禄；

【作者基本信息】 上海交通大学 ， 计算机技术， 2016， 硕士

【摘要】 在室内定位和室内地图创建中,获得实时的行走距离这个信息是非常重要的。但是,由于全球定位系统(GPS)在室内环境中会失灵,所以用最小的硬件条件来精确地估计室内的行走距离就非常地具有挑战性。总的说来,商品智能手机的惯性传感器的低端性能(例如加速度计和陀螺仪)、所需的额外的前提条件(例如一张室内地图)以及人的多样的步行状态(例如平速步行、快速步行和慢速步行)三个原因使得室内距离并不能容易且精确地获得。在本论文中,我们提出了一个轻量级的解决方案,叫做DiSen,来测量智能手机用户的实时的步行距离。在分析了大量的步行记录数据之后,我们发现人们具有相对一致的步行行为,即使他们在不同条件下改变他们的步行速度。并且,行走时的步长和步频之间的关系可以通过非线性的Sigmoid模型很好地估计。受这些深入观察的启示,我们首先设计了一个步子切割的方法来从原始的加速度传感器读数中获得可靠且准确的步频信息。接着,当某个用户在室外环境中行走时,智能手机同时收集加速度和GPS信息,我们用这两种信息来训练出一个专属于该用户的Sigmoid模型。最后,我们在室内将这个模型应用到步频信息上,就可以得到室内步行距离的测量值了。真实世界的实验结果显示,在不同的步行速度下,我们的系统,DiSen,的估计可以达到平均96%的准确度。更多还原

【Abstract】 Acquiring instant walking distance is desirable in indoor localization and map construction.However,due to the blackout of Global Positioning System(GPS)in indoor settings,to accurately estimate the indoor walking distance with minimum hardware requirement is very challenging.In a word,poor performance of commodity smartphones’ inertial sensors(such as accelerometer and gyroscope),additional prerequisite needed(such as an indoor map)and human’s diverse walking states(such as walking normally,fast and slowly),the indoor distance cannot be easily and accurately available.In this paper,we propose a lightweight scheme,called DiSen,to range the instant walking distance of smartphone users.After analysing the extensive walking trace data,we find that people have rather consistent walking behaviour even though they may change their walking speeds in different situations.Furthermore,the relationship between stride length and step frequency while walking can be well estimated using non-linear sigmoid model.Inspired by such insights,we first design a stride segmenting method to obtain reliable and accurate step frequency information from raw accelerometer readings.We then train a sigmoid model using acceleration and GPS information collected when a user walks in outdoor conditions and finally apply the model to indoor walking distance ranging.Real-world experiment results show that,in different walking speeds,DiSen can reach average distance estimation accuracy of 96%.更多还原