This paper presents an innovative approach to improving room prediction accuracy in autonomous indoor navigation systems by leveraging object differentiation and semantic graph-based methods. We significantly enhance prediction efficiency and accuracy by identifying unique objects and their spatial differences between rooms. The proposed method uses several trained modules, including feature extraction, decision tree classification, and real-time object detection, making the system lightweight and suitable for embedded devices. The overall results of the experiments show an exact algorithm capable of adapting to many indoor environments and improving the real-time navigation of autonomous systems with an improvement of 4–6% in accuracy across training, testing, and validation.