High Accurate and Efficient Image Retrieval Method Using Semantics for Visual Indoor Positioning

Abstract

Visual indoor positioning has a wide application because of its good positioning performance without additional hardware requirement. However, as the indoor scenes and complexity increase, the offline database will inevitably become large and the online retrieval time will also become long, which make visual indoor positioning unpractical. To solve this problem, we propose a Semantic and Content-Based Image Retrieval (SCBIR) method. By dividing the offline database into semantic databases with different semantic types, the retrieval scope of the image is reduced, and the retrieval time is reduced. First, we use the semantic segmentation method to detect the semantics. Then we divide different semantic scenes in terms of the image order and basic pattern of the semantics in the scene. Finally we use the images belonging to each different semantic scene to build a semantic database, so as to achieve online accurate and fast image retrieval. The experiment results indicate that the proposed method is suitable for large scale retrieval database, and it can reduce the retrieval time in the online stage on the premise of ensuring the accuracy of image retrieval that is critical for visual indoor positioning.