Research On Indoor Floor Localization And Topological Floorplan Construction Technology Via Crowdsourcing

With the advancement of the smart city process,intelligent instant delivery services require real-time and accurate indoor positioning technology to improve the delivery efficiency of couriers in complex multistory shopping malls and the system-human synergy experience in platform construction.Floor localization and indoor topological floorplan are two important components in the indoor positioning field.The former can provide floor/height information for 3D-based indoor navigation services(e.g.,delivery route planning),and the latter can display the floorplan to couriers in real-time.The organic combination of the two will contribute to realizing indoor 3D navigation services,further improving indoor delivery efficiency,and optimizing system order scheduling strategies.We argue that existing floor positioning methods cannot be deployed flexibly on a large scale due to label dependency and device dependency.This paper proposes a new self-evolving and user-transparent floor localization system named TransFloor,which is based on crowdsourcing instant delivery data(e.g.,order information and sensor data)without additional label investment and strict equipment constraints.TransFloor consists of an unsupervised barometer-based module-IOD-TKPD and an NLP-inspired Wi-Fi-based module-Wifi2Vec.And self-labeling is the perfect bridge between them to achieve fully label-free and deviceindependent floor localization.In addition,TransFloor,as a lightweight plug-in,is embedded into the platform without reconstructing the existing workflow.It has been deployed nationwide to adaptively provide real-time and accurate 3D/floor positioning services for all couriers.As for the problems of invalid PDR model inference and unstable trajectory matching performance from the existing methods,this paper proposes a novel indoor topological floorplan construction system named RoMatcher,which includes a TransFormer-based deep inertial odometry model called RoFormer and an automatic floorplan construction model based on geomagnetic similarity and trajectory similarity called DDTWAP.By combining them,RoMatcher can ignore the posture of the courier’s device and effectively construct the topological floorplan of indoor shopping malls and office buildings in crowdsourcing instant delivery scenarios,and provide couriers with 2D map information and indoor merchant locations.TransFloor is evaluated on the instant delivery platform involving 388 shopping malls containing 672,282 orders,7,390 couriers and 6,206 merchants.Experiments show that TransFloor can achieve an average accuracy of 94.61%,and perform good device heterogeneity and adaptive durability,outperforming existing state-of-the-art methods.As a case study,the platform can reduce the order scheduling error rate by 60%and the overdue delivery rate by 2.7%,reduce the arrival time of delivery personnel by 12.27 seconds on average,and improve delivery efficiency by 7.29%.RoMatcher is evaluated on some shopping malls and office buildings in Beijing.Experiments show that the trajectory inference performance of RoFormer on open-source data sets and crowdsourcing delivery data sets is better than the existing state-of-the-art models.68%of topological errors from RoMatcher in Beijing Huacai Center are less than 5.2m,and the errors on the 7th floor of Institute of Computing Technology are less than 2m,which verifies the feasibility of large-scale indoor topological floorplan construction deployment.