| Smartphone indoor positioning is one of the core technologies of Location Based Services(LBS)and is the key to achieve indoor navigation and related location services for mobile phones.Indoor positioning technologies have developed rapidly in recent years.When considering key factors such as accuracy,cost,coverage,complexity and applicability,existing indoor positioning technologies have their own applicability scenarios,which makes pervasive indoor positioning services for mobile phones more challenging.Visual features can be used as the main semantic information to help people understand the environment.Scene recognition not only extracts the geometric structure,texture information and semantic information of elements in buildings,but also obtains contextual information for related vision tasks such as object detection and motion recognition,and scene recognition technology fusing multiple source sensors provides a new way to solve indoor localization problems.At present,building map and scene recognition still have certain limitations when applied to mobile positioning,with the following three problems being the most critical:(1)insufficient semantic constraint information of building map;(2)immature matching positioning technology of building Map Location Anchor(MLA);(3)universal application problem of mobile phone indoor scene recognition positioning.To address the above problems,this paper proposes a mobile phone indoor scene recognition and localization method for building live maps,and the main research contents are as follows:(1)Research on building map construction method for mobile phone indoor scene recognition.This study takes BIM model as the basic data source,and builds a geocoded entity library of Map Location Anchor(MLA)based on the hybrid model of building map for semantic and geometric information expression in multiple scenes of building map,which can provide users with "immersive" real-world building map on one hand,and provide semantic anchor point constraint information for mobile phone positioning on the other.(2)Research on mobile phone indoor scene recognition and localization method for real building map.Based on the construction of the full element hierarchical classification of the scene for the hybrid model of the building map,the improved YOLOv5 s deep learning model on the mobile terminal is used to identify the element information of the location anchor point type in the building scene through the mobile phone video in real time during the pedestrian movement,and then the spatial location of the scene elements obtained based on the video recognition is spatially matched with the building Map Location Anchor(MLA)to achieve realtime mobile phone positioning.(3)Research on the empirical study of mobile phone indoor scene recognition localization method for building live map.Using the mobile phone indoor scene recognition and localization method under the semantic constraints of building map location anchors in this paper,MIPNS2.0,a mobile phone indoor location navigation system,is developed to verify the feasibility and practical value of the mobile phone indoor scene recognition and localization method for building live map in the process of universal scene application.The empirical results show that the system can effectively achieve high localization accuracy in building scenes with uniform distribution of MLA elements,and the recognition accuracy of 9 types of MLAs can reach 97.2%respectively,and the localization accuracy can eventually reach about 0.5 meters under the road network node constraints.This paper proposes a mobile phone indoor scene recognition and localization method and technical process for building live map,and shows through empirical evidence that the proposed mobile phone indoor positioning method has good usability and can meet the basic needs of users’ mobile phone indoor positioning and navigation,which provides an example with engineering application value for mobile phone indoor positioning service. |
PDF Full Text Request