Environment Description for Blind People

摘要

Visual processing is very efficient, letting people to use vision as the first approach to get information about environment. For blind people that information must be complemented with another very powerful data collection: sounds. In order to complement the white stick sounds, the prototype HOLO-TECH gathers and segments video images and produces specific sounds to acknowledge about potential hazards. The underlaying model is based on a set of Neural Networks coordinated by an Expert System to make it possible to react to any new event in real time. This paper presents an outline of the model, the project and a test set to evaluate one of the Neural Networks specialized to detect and evaluate faces and other objects like cars. The main contribution of this work is automate the selection model for proper combination of information, discarding unnecessary data and defining the minimum precision requirements to fulfill the current goal.