Deep Learning Coupled with Multiple-Model Adaptive Estimation for Indoor Localization/Tracking in IoT Applications******

该NSERC-ENGAGE项目旨在为工业合作伙伴dormakaba Canada Inc.开发/设计基于其低功耗蓝牙（BLE）设备的定位/跟踪解决方案。dormakaba Canada Inc.设计和制造门禁系统，如电子锁和读卡器，是加拿大和全球三大提供门禁和安全解决方案的公司之一，该解决方案适用于酒店、机场、医院和办公环境等各种场所。利用他们的设备进行室内定位/跟踪的优势在于，它避免了为他们的客户复制基础设施和设备。尽管dormakaba Canada Inc.拥有相应的基础设施，但目前dormakaba Canada Inc.还没有基于其支持BLE的设备开发本地化/跟踪解决方案。NSERC-ENGAGE项目旨在解决这一差距。dormakaba Canada Inc.正在寻找一种解决方案，利用安装在该空间内的具有ble功能的锁和读卡器以及移动设备，在限定的物理空间（例如建筑物）内对人进行微定位和跟踪。在这方面，最初将基于dormakaba Canada Inc.提供的BLE技术，在整个测试走廊安装传感器网络。将收集位置数据来训练深度神经学习架构（将研究卷积神经网络（cnn）和循环神经网络（rnn）的应用），以构建所选场地的接收信号强度指标（RSSI）地图。然后将训练好的模型耦合并集成到一个动态多模型估计框架中，以提高室内定位的精度

This NSERC-ENGAGE project aims to develop/design a localization/tracking solutions based on their Bluetooth Low Energy (BLE) enabled devices for dormakaba Canada Inc., the industrial partner. dormakaba Canada Inc., designs and manufactures access control systems such as electronic locks and readers and is one of the top three companies offering access control and security solutions in Canada and across the world incorporated in variety of venues such as hotels, airports, hospitals, and in the office environment. The advantage of leveraging their devices for indoor localization/tracking is that it avoids duplicating infrastructures and devices for their customers. Despite having the infrastructure in place, currently dormakaba Canada Inc. does not have localization/tracking solutions developed based on their BLE enabled devices. This NSERC-ENGAGE project aims to address this gap. In particular, dormakaba Canada Inc. is looking to find a solution to micro-locate and track a person within a delimited physical space (e.g. building) using their BLE-enabled locks and readers installed within this space in conjunction with a mobile device. In this regard, initially and based on the BLE technologies provided by dormakaba Canada Inc., a sensor network will be installed throughout a test corridor. Positional data will be collected to train deep neural learning architectures (application of convolutional neural networks (CNNs) and recurrent neural networks (RNNs) will be investigated) to construct the Received Signal Strength Indicator (RSSI) map of the selected venue. The trained models are then coupled and integrated within a dynamic multi-model estimation framework to improve the accuracy of indoor localization.**