Análise de desempenho da Google Cloud Vision API em leitura de textos provenientes de imagens naturais

Abstract

Great companies in the information technology market have been adopting techniques aimed to Artificial Intelligence and Computer Vision to solve daily problems such as facial recognition to unlock cell phones, detection of automobiles, people, translation of signs and warnings, etc. All of this, in real time or very close to this reality. Knowing that computer models are rapidly evolving and becoming increasingly robust and present in the real world, appears the curiosity to analyze these characteristics, such as robustness, speed, confidence and security of these tools. This work aims to analyze the accuracy of the Google Cloud Vision tool for detection and recognition of texts in natural images and find out what factors exist in the environment that we live and that can reduce the flexibility of a tool produced by a market giant.