Wi-Fi蓝牙融合定位方法研究与系统实现

发布时间：2018-09-14 15:26

【摘要】：随着人们活动的室内空间越来越庞大和复杂,兴趣点(Point of Interest,POI)越来越丰富,停车场、商场、机场等场所的定位和导航需求愈趋强烈。此外,精准位置营销、智能制造、机器人、无人医疗护理等行业也需要设备能够在室内识别特定对象的位置。这些都为室内定位技术(Indoor Positioning System,IPS)带来了巨大的机会。调查数据显示,人们在室内度过的时间占比达到80%以上。由于室内环境日趋复杂,空间越来越大,在停车场找车、逛商场找寻特定商品、联系走散的亲朋变得越来越难,这些问题推动了室内定位成为生活中的刚需。目前室内定位技术呈现百家争鸣的现象,却缺少一种定位技术能够在低成本的条件下满足位置服务的定位需求。超宽带定位、激光定位、红外定位、地磁定位等技术或需要专门的设备,或部署复杂、成本较高,难以实现大规模的推广。而基于Wi-Fi的指纹定位技术能直接利用场景中的现有设备,极大地减轻了定位系统部署的成本;蓝牙4.0的低功耗性、信号广覆盖性、低成本性也为这种新型的无线定位技术搭建好了舞台。然而基于Wi-Fi或者低功耗蓝牙的单模定位技术仍旧存在一定的局限性,由于室内环境高度复杂,信号的反射、衍射和多径效应都给基于无线信号的定位技术带来了很大困难;指纹定位技术在离线模型训练阶段大多需要相当丰富的训练样本才能学习出较好的定位模型,而更多的训练样本则意味着更长的模型学习时间和更大的数据采集工作量。针对上述问题,本文先后从指纹特征稳定性、模型训练的快速性和训练样本获取的便捷性以及最终的定位结果的有效性等方面着手,全文主要工作可分为以下三部分:1)基于互相关理论提出了一种融合特征提取方法。该方法首先对原始传感器信号采用高斯模型进行去噪处理,然后根据互相关理论进行互相关信息计算得到融合特征,最后再结合原始传感器特征得到组合特征。实验表明,该方法能有效提升指纹特征的稳定性和定位模型的精度和鲁棒性。2)提出了一种基于融合特征的半监督流形约束定位方法。首先引入超限学习机以提升模型的学习速度和泛化能力,然后引入半监督学习方法,采用拉普拉斯正则化来对模型进行流形约束,充分吸收无标记样本的数据特征,同时减少此类样本对模型的负面影响,从而增强模型的定位精度和鲁棒性。实验表明,半监督超限学习机的提出最多能将标定指纹的采集工作量减少90%,同时能将定位精度提升20%-30%。3)设计并实现了一种融合定位系统。该系统针对第三方应用开发者提供了一种快速集成室内定位功能的服务。开发者利用本定位系统的采集工具在定位场景中采集指纹后,再使用离线定位SDK(Software Development Kit)便可以快速体验室内定位功能。系统测试和轨迹重现结果表明我们设计的室内定位系统具有很好的实用性,也具备很好的商业价值。
[Abstract]:With the more and more large and complex indoor space of people's activities, the (Point of Interest,POI) is becoming more and more abundant, and the demand for positioning and navigation in parking lots, shopping malls, airports and other places is becoming more and more intense. In addition, industries such as precision location marketing, intelligent manufacturing, robotics, and unattended medical care also require equipment to identify the location of specific objects indoors. All these bring great opportunities for indoor positioning technology (Indoor Positioning System,IPS). Survey data show that people spend more than 80% of the time indoors. Because the indoor environment is becoming more and more complex and the space is more and more large, it is becoming more and more difficult to find cars in the parking lot, shopping malls to find specific goods, and to contact separated friends and relatives. These problems have pushed indoor positioning to become a rigid demand in daily life. At present, the indoor positioning technology presents the phenomenon of a hundred schools of thought, but the lack of a positioning technology can meet the location needs of location services under the condition of low cost. UWB positioning, laser positioning, infrared positioning, geomagnetic positioning and other technologies may require special equipment, or the deployment of complex, high cost, it is difficult to achieve large-scale promotion. The fingerprint location technology based on Wi-Fi can directly utilize the existing equipment in the scene, which greatly reduces the cost of location system deployment, the low power consumption of Bluetooth 4.0, the wide coverage of the signal, Low-cost also set the stage for this new wireless positioning technology. However, the single-mode localization technology based on Wi-Fi or low-power Bluetooth still has some limitations. Because of the high complexity of indoor environment, the reflection of signals, diffraction and multipath effect, the localization technology based on wireless signal is very difficult. Fingerprint localization technology in the off-line model training stage most of the training samples to learn a better location model, and more training samples mean longer model learning time and more data acquisition workload. In order to solve the above problems, this paper begins with the stability of fingerprint features, the rapidity of model training, the convenience of obtaining training samples, and the effectiveness of the final localization results. The main work of this paper can be divided into three parts: 1) A fusion feature extraction method based on cross-correlation theory is proposed. Firstly, the original sensor signal is de-noised by Gao Si model, then the fusion feature is obtained by cross-correlation information calculation based on the cross-correlation theory, and the combined feature is obtained by combining the original sensor feature. Experiments show that this method can effectively improve the stability of fingerprint features and the accuracy and robustness of the location model. 2) A semi-supervised manifold constrained location method based on fusion features is proposed. In order to improve the learning speed and generalization ability of the model, the over-limit learning machine is introduced first, and then the semi-supervised learning method is introduced. Laplacian regularization is used to constrain the model manifold, which fully absorbs the data features of unlabeled samples. At the same time, the negative effects of the samples on the model are reduced, so that the location accuracy and robustness of the model are enhanced. Experimental results show that the proposed semi-supervised over-limit learning machine can reduce the workload of fingerprint calibration by 90% at most, and at the same time, it can improve the positioning accuracy by 20- 30.3) and design and implement a fusion positioning system. The system provides a fast integrated indoor location service for third party application developers. The developer uses the acquisition tool of the positioning system to collect fingerprints in the location scene, and then use the off-line positioning SDK (Software Development Kit) to quickly experience the indoor positioning function. The results of system test and trajectory reconstruction show that the indoor positioning system designed by us has good practicability and commercial value.
【学位授予单位】：湘潭大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN92

【参考文献】