Semantic and Context Based Image Retrieval Method Using a Single Image Sensor for Visual Indoor Positioning

Visual indoor positioning method plays an important role in indoor positioning since it does not require pre-deployed facilities. There are two problems in the traditional methods. On the one hand, the image retrieval is very time-consuming due to large number of images in the offline database. On the other hand, the positioning accuracy is also not ideal since there are many mismatching pairs between query image and matching image. In this paper, a novel visual indoor positioning method based on semantics is proposed to address the problem of low efficiency and poor accuracy. Firstly, we select indoor common and representative infrastructural objects as semantics to avoid decline of retrieval accuracy caused by the change of indoor environment. Meanwhile, we employ semantic extraction and classification for the collected images with common and representative infrastructural objects to establish the semantic based offline database. In addition, the semantic and context based image retrieval method is proposed to obtain the optimal matching image. Finally, the semantic constraint based feature point selection method is adopted to estimate the user’s position. The simulation results show that our proposed method can improve the efficiency and accuracy of visual positioning.