This project aims to develop a cost-effective, real-time assistive vision system that converts live camera video into spoken feedback for visually impaired users. Designed as a wrist-worn, non-invasive solution, the system enables reliable object localization within the user’s immediate surroundings. It is built around a low-power ESP32-S3 microcontroller with an integrated Wi-Fi stack and a compact OV2640 camera module. Lightweight deep-learning models, such as YOLOv8n, are employed for efficient object detection, while Depth Anything V2 provides monocular depth estimation, optimized for indoor environments and relative depth accuracy. Video data is streamed over TCP/IP via Wi-Fi to a local computer for external AI inference, enabling high frame rates while keeping the wearable hardware energy efficient. A multi-threaded software architecture synchronizes detection and depth outputs, which are then translated into spoken descriptions using Text-to-Speech. Overall, the system balances affordability, portability, battery life, and real-time performance to deliver practical assistive feedback.