Researches On Vision-based Indoor Positioning And Navigation For Pedestrian In Large Indoor Scenes Based On Ubiquitous Objects

"Indoor ubiquitous objects" refers to objects which distribute in general or specific indoor scenes widely and commonly.They have several typical features: ubiquity in indoor scenes,Non-landmark functionality(such objects have their own functions instead of landmark functions),Visual saliency(easily captured by moving pedestrian),Steady and long-term fixity.These ubiquitous objects in indoor scenes contain abundant geographic location information,which can be excavated and applied to indoor location services.Typical indoor ubiquitous objects include doors,windows,stairs and shop brands in shopping malls,etc.Researches on spatial cognitive indicated that the visual attraction of landmarks is not the main factor considered by pedestrians when judging the path of indoor,but the functional landmarks(such as doors,stairs,etc.)are often more concerned and thus are more suitable for guiding pedestrians to move.These features make indoor ubiquitous objects good location references for vision-based indoor positioning and navigation methods.Based on ideas above,this paper researched and designed vision-based positioning and navigation methods for large indoor scenes,and its main contents include: 1.The paper proposed a visual indoor positioning method based on general ubiquitous objects in large indoor scenes.The method firstly detects and recognizes ubiquitous objects in images shot by users via deep learning methods,then gets the pixel coordinates as well as corresponding space coordinates of control points at targets,finally calculates and outputs position of users.As it avoids the global digital modeling for indoor environment,and it is not necessary to construct any infrastructure or makes any changes for the indoor environments,our method lowers the requirements of data acquisition,processing,and storage.The experiment in a large art museum proved that the proposed method could realize 1-meter positioning accuracy within 40 meters' range.2.For a specific kind of ubiquitous objects,i.e.scene texts in shopping malls especially the store names,we proposed a visual navigation method based on scene text to navigate pedestrians in large shopping malls.Without building a 2D image database or a 3D model of indoor scenes,our method simply detects and recognizes scene texts in shopping malls(like shop names),and relates them with the plain map of the shopping mall to estimate the location of pedestrians.Since users input their destination,the system could find the shortest path and display it for navigation.Compared with traditional visual indoor positioning strategies with 2D or 3D databases,this approach only need to build shapefiles of shopping mall plain maps,which greatly reduces the resources of data acquisition and storage.Besides,due to the strong connection between shop names in indoor scenes and plain maps,together with the robustness of algorithm for scene OCR,the proposed method has relatively stronger adaptability to the change of shopping mall environment.The experiment in a large shopping mall proved the feasibility of the method.