Design and Implementation of an Intelligent Navigation System for the Visually Impaired Based on Multimodal Perception Fusion

The increasing demand for independent mobility among visually impaired individuals, especially in complex urban environments, has highlighted the limitations of traditional assistive tools such as white canes and tactile paving. These conventional methods offer only basic spatial awareness and lack the capacity to adapt to dynamic obstacles or rapidly changing surroundings. In response, this study presents the design and development of a lightweight, embedded navigation system based on multimodal perception fusion. The system integrates visual, ultrasonic, and inertial sensing modules, orchestrated through an ST Microelectronics 32-bit Microcontroller (STM32), and employs compact deep learning models to enable real-time obstacle detection, semantic scene understanding, and user behavior monitoring. Additionally, it incorporates BeiDou satellite positioning and Global System for Mobile Communications (GSM) communication for continuous location tracking and emergency alerts, while a voice interaction interface ensures effective user feedback. Fall detection mechanisms are also embedded to enhance safety in unexpected situations. Experimental evaluation in real-world urban settings demonstrates that the system maintains stable performance, achieves positioning accuracy within 8 meters, and ensures a reliable runtime of approximately 8 hours on a single battery charge. These results validate the feasibility and effectiveness of integrating embedded AI and sensor fusion technologies to improve autonomous navigation capabilities for visually impaired users.