The Construction Of Indoor Semantic Mapping Based On Non-cooperative Environment

Traditionally,the indoor semantic map is constructed on the basis of existed deployment environment.Nevertheless,under most circumstances,plenty of environments exist without specific deployment requirement.Also,in a lot of strange and uncertain scenes,the corresponding indoor semantic map cannot be obtained.The demand of specific deployment environment decreases the available areas,especially for those with lots of application like malls and with special functions like military areas.Therefore,this paper focuses on an available method for building indoor semantic maps which is not limited to the requirements of devices in specific scenarios.At the same time,the basic data of this approach is supposed to make full use of the large number of location information generated by the movement of people in the scene to provide the data basis for the construction of our indoor semantic map.It is worth noting that in recent years,with the widespread popularity of smart phones,it has become an indispensable part of people's lives.The built-in functional sensors(accelerometers,gyroscopes,magnetic sensors,etc.)provide a means to obtain the location data generated by pedestrians moving in different scenarios.However,it is a non-subjective behavior for users to generate a large number of location data when they move in the scene.Under this premise,we call it “non-cooperative” state.At the same time,the scenario environment does not exist for subjective purposes.Our construction work is based on the existing environment and “non-cooperative” implementation is not to set specific scenarios for realization.Based on this,our paper proposes an indoor semantic map construction method under non-cooperative environment.Rooms and corridors in indoor semantic maps are the basic framework and play an important role in the process of building the overall map.Therefore,this paper focuses on the room and corridors.Android smartphone is used to collect random location data generated during pedestrian walking.Convolutional neural network is used to identify and obtain scene information(room or corridor)corresponding to each trajectory location point,so as to provide semantic information for subsequent map construction.Secondly,Delaunay triangulation network is set up after sliding window processing for data points containing semantic information to ensure the complete shape and layout of the scene area to be built.After that,we propose the construction of indoor semantic map model and spatial topological semantic constraint model(angle constraint modification model and location relationship modification model).On the basis of the triangle network that has been constructed,the corresponding semantic model is constructed regularly.Then,according to the relevant modification technology,we can further modify and improve the regularized scene and finally get the indoor semantic map we need.