Blind Assistance System for Easy Access of Home Appliances

The purpose of this research is to design, develop, and assess a blind assistance system that will allow blind people to safely and independently handle household appliances. It's a well-known fact that the estimated number of visually impaired individuals worldwide is approximately 285 million, which is roughly equivalent to 20% of the Indian population. These individuals face consistent challenges in terms of accessibility, especially when they are on their own. Often, they rely on assistance from others even for their basic day-to-day needs. Consequently, addressing this issue through technological means becomes a vital and much-needed endeavour. Hence, this Smart Assistance System is designed to empower visually impaired individuals, enabling them to independently identify, classify, and interact with common day-to-day home appliances in real time. In this paper, three CNN algorithms are compared in this study. Accuracy, precision, recall, F1 score, and hamming loss on the validation image are used as metrics to evaluate the performance of the model. The F1 scores for the custom architecture CNN, Mo-bileNetv2, and Yolo model were 0.43, 0.63, and 0.24, in that order. This paper suggests categorizing bounding box buttons using YoloV8. Many objects can be instantly recognized by the sophisticated YOLOv8 model. The described approach uses a convolutional neural network to classify the object of interest. In the testing image of home appliance buttons, the solution outperforms the state-of-the-art algorithms with 95% classification accuracy.