Assistive Tech For Blind

The Assistive Tech For Blind project operates as an edge-based artificial intelligence system that transforms visual data into auditory feedback. At the core of the system is the Sipeed Maixduino Kit for RISC-V, which serves as the processing unit running a custom-trained object detection model. The attached camera module continuously captures the user's surrounding environment in real-time. This video feed is processed locally on the edge hardware using a lightweight neural network architecture derived from Ultralytics YOLOv5, which identifies and classifies physical obstacles and 3D objects in the path of the user. Once an object is detected, the Maixduino processes the spatial and categorical data and triggers an audio output signal. This signal is sent directly to a speaker or wired headphones, translating the visual detections into spoken words or distinct audio alerts. The onboard LCD display can optionally provide a visual preview of the camera feed and bounding boxes during debugging. Software development and model training are performed externally using a computer and Google Colab, after which the compiled firmware and model weights are flashed onto the board via a USB Type-C cable using tools like Kflash_gui and MaixPy IDE.