Automatic Identification of Braille Blocks by Neural Network Using Multi-Channel Pressure Sensor Array

Abstract

In recent years, the number of visually impaired people in Japan has exceeded 300,000 including those with low vision, and accidental falls on the station platform involving them have not been eliminated. Persons having acquired visual impairment make up one third of all cases of blindness in Japan. It is known that they often cannot walk alone with only a white cane or guide dog. The main cause of platform accidents was misidentification of braille blocks. Therefore, it was necessary to develop an auxiliary device for accurately identifying braille blocks that the acquired visually impaired could also use easily. In this research, we developed an automatic identification system for braille blocks using foot pressure data acquired by a multi-channel pressure sensor array. First, we devised a new foot pressure data acquisition device using a multi-channel pressure sensor array. Our proposed device had excellent features such as being light weight, low cost, and easy to extend to multi-channel. Second, in order to accurately identify the braille blocks, the foot pressure data acquired under various conditions was learned by neural network, and identification performance evaluated. As a result of the experiment, the braille blocks could be identified with a high rate of at least 98% accuracy by neural network, with a very simple structure of an input layer (16 neurons), a hidden layer (5 neurons), and an output layer (4 neurons).