Using Image-based positioning for seamless localization in cultural heritage setting

This study presents the development and evaluation of an image-based positioning system for a mobile museum visitors’ guide system comprising two components: An Android application and a backend service. The application identifies visitors’ location by capturing and sending images to the server, which then determines their location within the museum. The server maintains a comprehensive dataset of the museum's points of interest (POIs), and content about them. The content was created automatically, using large language model (LLM) and corrected by museum staff, who can also upload videos and descriptive information for each POI via the application. The image-based indoor positioning solution uses a deep learning-based model for representing an image as a vector of features. This approach enables the system to simply calculate distances between vectors and ultimately determine the similarity between them, allowing for accurate POI identification. A user study aimed at evaluating users' perception of the systems' accuracy and ease of use was conducted at the Hecht Museum, where participants used the developed application and subsequently completed a System Usability Scale (SUS) questionnaire, along with other open-ended questions. The high scores and the highly positive feedback obtained indicate an overall excellent usability experience, especially with respect to the accuracy and speed of POI identification. The feedback also provided insights into areas where our solution can be enhanced and further developed.