Development of indoor localization system using a mobile data acquisition platform and BoW image matching

Recently, an increasing interest in the Location-Based Services (LBS) has led to the needs for studies on indoor localization. Therefore, this study aims to propose an image-based indoor localization system, which can improve the practicality and efficiency for wider applications. In order to construct reference imagery database, an optical camera and a 2D laser scanner mounted on a mobile platform were utilized and the data acquired was registered to the real world coordinate system. The localization procedure was then conducted based on Bag of Words (BoW) algorithm which matches the images by evaluating the similarities of the histograms produced from them. In addition, the experiments to verify the optimal system design were carried out. Compared to manual construction, the mobile platform-based data acquisition and location registration can be carried out in an easier manner. The proposed localization method provided more than 80% of localization accuracy for the query images taken in various locations and points of view. The system proposed herein is a practical, efficient, and accurate means of constructing the reference images and determining the location of a user in the indoor environment. It was also found that the size of codebook and the number of reference images should be determined while considering the computational time and the purpose of the application. Reducing the difference of the spatial resolution between the reference and the query images also provided more accurate results.