Research On Visual Indoor Positioning Algorithm Based On Simultaneous Localization And Mapping

In recent years,with the popularity of camera in terminal device and people's desire for entertainment life,the location-based services are concerned by lots of researchers and companies,such as the intelligent recommendation,navigation,AR and so on.So how to obtain the user's position is the basis of all subsequent services.In the outdoor,it can use mature GPS positioning method.But in the complex and changeable indoor environment,the positioning accuracy of GPS is insufficient.Therefore,a variety of indoor positioning technology is proposed.It's worth nothing that vision-based indoor positioning has become the first choice for many indoor applications because of the advantages of high precision and no additional equipment.Based on the theory of digital image processing,deep learning,multi-view geometryand three-dimensional reconstruction,this paper combines the semantic recognition,simultaneous localization and mapping(SLAM)technology and vision-based indoor positioning algorithm.Firstly,this paper discusses the research status of slam and vision-based positioning at home and abroad.Secondly,this paper analyzes the basic theory of vision-based indoor positioning algorithm and slam technology.In a general way,vision-based indoor positioning mainly consists of three key research parts:the establishment of offline database,image retrieval and positioning algorithm.This paper mainly studies on some shortcomings of traditional algorithms about two aspects:the establishment of offline database and image retrieval.(1)Aiming at the problem that the traditional SLAM algorithm is not robust indynamic environment,this paper proposes a SLAM algorithm which integrates semantic and geometric information.At first,the semantic information of the input image is recognized by the semantic detection network,and then the dynamic points in the image are detected by the multi-view geometric relationship in the semantic region.At last,the front-end combines the semantic and geometric information to optimize the pose.This method can effectively enhance the robustness of slam system in the dynamic environment,and improve the accuracy of camera pose that from the global world coordinate to the local camera coordinate.(2)Aiming at the shortcomings of low precision and large database capacity of 2D image-based positioning algorithm,and the disadvantages of low retrieval efficiency of 3D model-based positioning algorithm,a vision positioning method based on slam is proposed.In the offline stage,the offline database is established by combining 2D images and 3D point cloud image obtained by slam.In the online stage,two-step matching method is used to obtain matching 2D-3D matching pairs,and user position is solved based on it.This method can effectively improve the positioning accuracy as well as the positioning efficiency under the premise of ensuring the small database capacity.The visual indoor positioning method based on SLAM proposed in this paper has broad application prospects.Among them,the SLAM algorithm,which combines semantic and geometric information,is more robust in the dynamic environment.At the same time,the establishment of offline database based on SLAM technology effectively reduces the time cost of the establishment of the traditional visual indoor positioning algorithm.The proposed method can effectively improve the positioning accuracy and efficiency on the premise of ensuring that the capacity of offline database is small.