Echo Vision: AI Voice Assisted Navigation for Visually Impaired

Echo Vision is a mobile assistive navigation system that uses augmented reality (AR) and edge computing to guide visually impaired users. We use ARCore for real-time camera pose tracking on smartphones that support it. If a device does not support ARCore, we use ORB-SLAM3. The system constantly checks if the environment is indoor or outdoor using a deep scene recognition model (Places365). This information helps select the correct object detection method. A YOLOv8 model is used for object detection indoors, and an SSD Mobile Net V3 detector is used outdoors for important obstacles. The phone sends video frames to a local server over Wi-Fi, and the server returns identified objects with their locations and distance ranges (for example, 0–1 m or 1–1.5 m). A sliding memory module on the server combines recent detections and creates periodic voice prompts that summarize nearby obstacles. This avoids too many repeated alerts. The client application shows visual overlays of the environment and obstacles and gives voice warnings through Android TextToSpeech, with vibration alerts for objects closer than 1 m. Tests in hallways and outdoor paths show that EchoVision can adjust its frame rate based on changes in the scene. This helps manage network use. This paper describes the architecture, implementation, and performance of Echo Vision. It also explains the current features and future steps for an edge-cloud assistive platform.