空间认知增强高可用高精度手机室内定位

High-precision indoor positioning is one of common key technologies for indoor navigation, location based services (LBS) and a variety of disruptive technologies of artificial intelligence (AI). This project aims to a high-usability smartphone indoor positioning solution of centimeter-level accuracy by fusing heterogeneous observables of multiple ubiquitous positioning sources built in a smartphone, including visual sensors, inertial measurement units (IMU) and radio-frequency (RF) signals. The high-usability and high-precision solution is achieved by exploiting machine learning algorithms and the latest advances of indoor mobile mapping technology and smartphone emerging in-built low-cost sensors. The research plan would break through visual absolute positioning technology by matching spatial features such as point, line, plane and cube with high-precision indoor realistic maps, including LiDAR point cloud and panorama images, which are collected by mobile mapping. By such, complicated indoor topology becomes meaningful features rather than error sources like in traditional positioning methods, and more complicated indoor structures are, richer spatial features are, and it is more advantageous for positioning. High-precision visual positioning is augmented by high-availability ubiquitous positioning solution that fuses smartphone RF signals and pedestrian dead reckoning (PDR), and heterogeneous observables are integrated using unscented Kalman filter (UKF) through a two-stage hybrid positioning scheme “coarse positioning to high-precision positioning”. Finally, this project will result in a high-usability smartphone indoor positioning solution of centimeter-level accuracy. The proposed solution provides absolute positioning under a global coordinate system, and therefore it is easy to integrate with GNSS positioning to achieve indoor/outdoor seamless positioning. This project is able to provide high-precision location of indoor pedestrians and mobile things, and the outcomes can be applied extensively, e.g., for smartphone intelligent LBS, robot localization and virtual reality (VR) applications.

高精度室内定位是室内导航、位置服务和多种人工智能颠覆性技术不可或缺的共性关键技术。当前，高可用低成本手机室内定位的精度是2-5米，分米级或更高精度定位面临室内空间拓扑结构复杂、物理信道干扰源多和室内运动情景多样性等难题。针对上述难题，本项目利用室内移动制图技术和手机新传感器等最新进展，融合点云和图像数据，从高精度三维实景室内地图提取点线面体等特征，将室内空间环境从传统方法中的误差源转变成有用的定位特征信号，室内空间环境越复杂，定位特征越丰富，越有利于定位精度；发展手机内置视觉传感器、运动传感器和泛在信号等多种定位源异构数据智能融合定位的理论方法，提出“粗定位→精定位”二级定位体制，形成在复杂多变的室内运动情景中高可用、开机即得的厘米级精度手机室内定位系统性体系。本课题的研究将为室内行人和移动平台等提供室内高精度位置信息，研究成果能广泛应用于手机智能位置服务、机器人定位导航和虚拟现实等。

高精度室内定位是室内导航、位置服务和多种人工智能颠覆性技术不可或缺的共性关键技术。高可用低成本手机室内定位的精度通常是2-5米，优于1米精度定位面临室内空间拓扑结构复杂、物理信道干扰源多、多类定位源异构性、手机设备差异性和室内运动情景多样性等难题。针对上述难题，本项目利用室内移动制图技术和手机新传感器等最新进展，将室内空间环境从传统方法中的误差源转变成有用的定位特征信号，室内空间环境越复杂，定位特征越丰富，越有利于定位精度。本项目主要研究多传感器融合室内移动扫描原型系统、视觉匹配全局定位技术、手机运动情景理解和多模态PDR航迹推算技术、基于手机的泛在信号室内定位方法、基于环境感知和运动情景理解的视觉/WiFi/PDR/蓝牙和地图等全源数据融合室内定位方法。建立基于环境感知增强的“高精基准控制、多源智能融合”多源混合高精度室内定位理论与方法体系，重点突破了室内视觉/LiDAR融合SLAM技术、多源泛在信号指纹移动采集和基于感知压缩理论的稀疏指纹库建库与定位方法、基于场景识别与运动情景理解的异构多源自适应融合定位等方法，提出“粗定位→精定位”二级定位体制，形成在复杂多变的室内运动情景中高可用高精度定位方法和关键技术。项目在多传感器融合移动扫描、视觉定位、基于手机的泛在定位和多传感器融合定位等理论和方法领域发表高水平论文28篇，其中SCI收录论文23篇。在知识产权方面，项目在相机与激光雷达标定、泛在信号定位、多传感器融合等领域申请专利7项，其中授权5项，获得软件著作权6项，均超额完成项目预期成果目标。项目成果包括多传感器融合室内移动扫描原型系统、手持激光雷达扫描室内定位仪、手机室内定位引擎和AR导航软件等，部分成果已应用于医院室内智能导医、室内AR导航等位置服务，并与华为公司合作转化应用于华为河图系统开发。本项目为室内行人和移动平台等提供室内一种高精度定位解决方案，成果能广泛应用于手机智能位置服务、机器人定位导航和虚拟现实等。