Vision-Based Real-Time Object Detection and Voice Alert for Blind Assistance System

Blindness is a very common and unendurable disability among many disabilities. About 285 million people are visually impaired in the world, as per the WHO. Visually impaired people often face various challenges in their daily lives, particularly when it comes to navigating unfamiliar locations and recognizing unfamiliar objects in their surroundings. State-of-the-art computer vision technologies can be used for assisting visually impaired people. Real-time object detection is a difficult operation as it requires more computing power to recognize objects in real time. SSD (Single Shot multi-box Detection) is a quicker detection method that is suitable for real-time object detection. A lightweight network model is required as the prototype device is tested with a Raspberry Pi microprocessor. Therefore, MobileNetV2 is incorporated with the SSD model to increase the accuracy level of detecting real-time household objects. The developed portable prototype comprises Raspberry Pi 3 model B, a USB camera, earphones, a distance sensor, and rechargeable batteries. The objects captured by the camera are classified by the object detection model and the audio feedback is given to the user through the earphones.