Abstract : The navigation system plays a crucial role in determining the pedestrian's position and providing the optimal route to a specific destination. While outdoor navigation systems rely on GPS signals and line of sight with orbiting satellites for localization, these methods are not applicable for indoor environments. In indoor settings, various technologies such as Infrared (IR), Wi-Fi, and RFID have been employed for localization. However, this research project focuses on computer vision-based techniques for indoor positioning. To achieve indoor positioning, a visual indoor positioning system utilizing a CNN-Based image retrieval method has been developed. The system's database comprises images representing each signboard within the room or store, along with their corresponding CNN features. Furthermore, the system provides absolute coordinates with respect to a given local coordinate system and scene labels. In this study, we propose the utilization of the pre-trained ResNet-18 deep learning model to determine the pedestrian's position and destination. The user inputs their desired destination, and the system employs Dijkstra's algorithm to calculate the optimal route. The ResNet-18 model has demonstrated an impressive accuracy of 92% in preliminary testing. This research project aims to enhance indoor navigation by leveraging computer vision techniques and deep learning models. By providing accurate positioning and optimal routes, the system holds great potential for improving navigation experiences within indoor environments. The findings of this study contribute to the growing field of indoor localization and pave the way for further advancements in this domain.


PDF | DOI: 10.17148/IARJSET.2023.10561

Open chat