基于众包数据的室内定位方法和平台

发布时间：2018-07-23 16:47

【摘要】：随着WLAN的普及,基于Wi-Fi的室内定位方法逐渐成为研究与应用的热点。虽然,其中基于位置指纹的定位算法研究相对广泛,应用效果较好,然而现有的指纹定位方法或系统仍存在以下3个问题:(1)离线阶段的数据标定和定位模型的训练需要耗费大量人力物力,以及时间消耗,使系统很难得到实际应用;(2)真实环境中WLAN信号波动呈现高动态性,采集的数据存在显著的时效性,无法提供长时间的有效定位保证;(3)实际环境中AP设备变动频繁,导致训练数据与定位数据特征维度不等长,造成模型失效。针对上述问题,本文提出了一种基于众包数据的模型更新方法,通过不断融合增量数据,使定位模型保持实时有效。该方法主要包括半监督极速学习机(SELM)、具有时效机制的增量式定位方法(TMELM)和特征自适应的在线极速学习机(FA-OSELM)3部分。基于上述方法,本文设计并实现了基于众包数据的室内定位平台系统。实际应用表明,本文提出的方法能够显著降低模型训练阶段的数据采集工作量,有效提升模型训练速度,并且长时间保持较高的定位精度。
[Abstract]:With the popularity of WLAN, indoor positioning method based on Wi-Fi has become a hot spot in research and application. Although, the research of location fingerprint based location algorithm is relatively extensive, and the effect of application is good. However, the existing fingerprint localization methods or systems still have the following three problems: (1) the training of off-line data calibration and location model needs a lot of manpower and material resources, as well as time consumption. It makes the system difficult to be applied in practice; (2) the fluctuation of WLAN signal in real environment is highly dynamic, and the collected data has significant timeliness, which can not provide effective positioning guarantee for a long time; (3) AP equipment changes frequently in real environment. The result is that the feature dimension of training data and location data is unequal, which results in the failure of the model. To solve the above problems, this paper proposes a model updating method based on crowdsourcing data, which can keep the location model real-time and effective by continuously merging incremental data. This method mainly consists of three parts: (TMELM), an incremental localization method with time-dependent mechanism, and FA-OSELM (feature adaptive online extreme learning machine), which is a semi-supervised extreme learning machine (SELM),). Based on the above methods, this paper designs and implements the indoor positioning platform system based on crowdsourcing data. The practical application shows that the proposed method can significantly reduce the workload of data acquisition in the training stage of the model, effectively improve the training speed of the model, and maintain a high positioning accuracy for a long time.
【作者单位】：湘潭大学信息工程学院;中国科学院计算技术研究所;北京市移动计算与新型终端重点实验室;中国科学院大学;
【基金】：国家自然科学基金项目(61572004、61472399) 中国科学院科研装备研制项目“面向可穿戴行为感知的精准模型测试仪研制”(YZ201527)
【分类号】：TN92

【相似文献】