基于自动学习的深度置信网络及多源信息融合的复杂环境下室内定位关键技术研究

With the development of Internet of things, people put forward higher requirements for accuracy, continuity and stability of indoor wireless positioning. However, because the indoor positioning technology exists some influence factors such as NLOS (non-line of sight), complex electromagnetic environment and changeable space layout, there is no acceptable technology can dominate the high-precision indoor positioning so far. This project will focus on the study of geometric high precision positioning in the LOS (line of sight) environment based on UWB (ultra wideband), the study of fingerprint positioning in the NLOS environment based on the deep learning of the fusion of UWB energy spectrum characteristics vector and geomagnetic intensity signal, and the study of DR (dead reckoning) navigation based on accelerometer and gyroscope and the simultaneous localization and mapping(VSLAM) in the moving environment based on the extended Kalman filter(DR-VSLAM). At the same time, based on the NLOS recognition algorithm, in LOS environment, the positioning will be realized using the fusion of UWB geometric positioning and DR-VSLAM and the training samples of the geomagnetic fingerprint will be updated in this case; in NLOS environment, the positioning will be realized using the fusion of fingerprint positioning and DR-VSLAM and the energy spectrum training samples of the UWB fingerprint will be updated in this case. According to these methods, the high accuracy, continuous and stable indoor wireless location will be achieved under NLOS and complicated electromagnetic environment based on the automatic learning deep belief nets and multi-source information fusion.

随着物联网的发展，人们对室内无线定位的精确性、连续性和稳定性提出了更高的要求，但是由于室内定位技术存在非视距、电磁环境复杂、空间布局多变等影响因素，迄今为止尚没有一种能够主导的高精度室内无线定位技术。课题将重点研究在视距环境下基于UWB（超宽带）的高精度几何定位，在非视距环境下基于深度置信网络的UWB能量谱特征向量和地磁强度融合的信号指纹定位，在运动环境下基于扩展卡尔曼滤波实现加速度计、陀螺仪的航位推算(DR)与基于视觉的即时定位与地图构建(VSLAM)的融合(DR-VSLAM)；同时基于非视距识别算法实现在视距状态下用UWB几何定位与DR-VSLAM的融合定位及地磁指纹训练样本的实时更新，在非视距状态下用信号指纹定位与DR-VSLAM的融合定位及对UWB能量谱样本的实时更新。最终项目将实现非视距、复杂环境下深度置信网络的自动学习及多源信息的融合，从而实现高精度、连续、稳定的室内无线定位。

随着物联网的应用越来越广，对室内定位的要求越来越高，减小室内定位中普遍存在的电磁环境复杂、非视距、布局多变等不利因素的影响，提高室内定位的质量，实现稳定的高精度室内定位方法成为亟待解决的问题。. 项目研究了视距环境下基于标准差迭代的测距方法、基于信道均衡技术的相关接收方法和基于锁相环技术的相关接收方法；非视距环境下通过小波包分解构造能量谱特征向量的方法、基于深度置信网络的指纹定位方法以及将自适应遗传算法与神经网络相结合的定位方法；运动状态下将改进的粒子滤波用于多源信息融合定位的方法、以及将视觉即时定位与地图构建（Simultaneous Localization and Mapping，SLAM）与惯性测量单元（Inertial Measurement Unit,IMU）融合定位的方法。最终实现了高精度、连续、稳定的室内无线定位方法。. 通过实验对该研究进行了评估，在定位的准确性、连续性等方面都取得了很好的效果，在多方面达到了预期目标，解决了室内定位中的关键技术问题。研究成果还可以用于无线电射频识别（Radio Frequency Identification，RFID）等其它无线定位技术中，并且取得了很好的定位结果。. 相关研究成果共发表论文15篇，其中SCI/EI检索论文7篇、中文核心论文3篇，申请发明专利5个，培养硕士研究生5名。