基于移动对象轨迹的室内导航网络构建方法

室内导航网络是行人导航、信息推荐和商业分析的基础。传统人工测绘或半自动提取的室内三维导航网络无法满足复杂室内空间结构高频变化需求。随着室内定位技术的不断发展,室内移动对象轨迹数据爆发式增长,为室内导航网络快速构建与变化监测更新提供了可能。本文提出一种基于移动对象轨迹的室内导航网络构建方法,在基于ST-DBSCAN的轨迹简化预处理基础上,提出了室内轨迹自适应栅格化算法,减弱栅格图像分辨率对导航网络提取的影响,有效避免廊道轨迹密度差异造成的导航网络拓扑连通失效,并通过CFSFDP自适应聚类算法自动识别楼层之间连通点,实现室内导航网络的快速构建。实验数据来源于上海图聚智能科技股份有限公司提供的某商城真实的室内移动对象轨迹数据,实验结果表明,与普适栅格化方法相比,本文提出的方法将导航网络构建准确率平均提高2.43%,拓扑正确度提高12.8%。

A Method for Constructing Indoor Navigation Networks based on Moving Object Trajectory

The indoor navigation network is the basis for pedestrian navigation, information recommendation, and business analysis. The traditional method of manual mapping or semiautomatic extraction of three-dimensional indoor navigation network cannot meet the requirement of high-frequency change of complex indoor space structures. With the continuous development of indoor positioning technology, there is an explosion of trajectory data of indoor moving objects, which provides a possibility for rapid construction and change monitoring of indoor navigation networks. This paper proposes a method of crowdsourcing construction of indoor navigation network based on the trajectory of moving objects. Based on trajectory simplification preprocessing using ST-DBSCAN, an indoor trajectory adaptive rasterization algorithm is proposed to reduce the influence of raster image resolution on the extraction of navigation networks. This approach effectively avoids the failure of navigation networks' topological connection that is caused by the difference of track trajectory density. Moreover, it automatically identifies the connection points between floors by the CFSFDP adaptive clustering algorithm to realize the rapid construction of indoor navigation networks. The experimental data is derived from the real indoor moving object trajectory data provided by Shanghai Palmap Science & Technology Co., Ltd. The experimental results show that, compared with the universal rasterization method, the proposed method improves the accuracy of navigation network construction by an average of 2.43% and improves the accuracy of topology by 12.8%.