Indoor localization via multi-view images and videos

摘要

Indoor localization systems are applicable to labyrinth-like environments where mobile and robotic operators require precise direction and location. Existing indoor localization systems require additional equipment, such as WiFi receivers and distributed beacons to capture the necessary information to accurately calculate direction and location. Image-based localization is mostly used in outdoor environments to compensate for the deficiency of weak GPS signals in large building areas. This paper presents a new indoor localization method that significantly reduces the dependency of external hardware resource by utilizing multiple-view configuration. The proposed algorithm initially extracts the self-similarity matrix features to represent the video that captures the user’s spatio-temporal information. We then perform the image and video retrieval based on the trained multi-task classifiers to determine scene location and orientation. Our method generates accurate location and orientation estimation.