This study presents the development and evaluation of a real-time assistive navigation system integrating YOLOv8 and YOLOv11 object detection models with the BLIP image captioning framework. Indoor object images across ten classes were collected, annotated, and used to train both models under identical conditions. Model evaluation metrics demonstrated that YOLOv11 achieved superior precision, recall, and mean Average Precision compared to YOLOv8, making it the preferred choice for deployment.

The selected YOLOv11 model was integrated with BLIP to generate contextual natural language descriptions, enhancing navigation feedback for visually impaired users. An Android application was developed to facilitate real-time navigation through voice input, GPS-based routing, and text-to-speech guidance, supported by a FastAPI backend for image processing and caption generation. The integrated system was tested in real-world scenarios, confirming its potential for enhancing user mobility and situational awareness.