AN APPROACH BASED ON CNN TO RESIDENTIAL ENVIRONMENT CLASSIFICATION FOCUSED ON REAL ESTATE BUSINESS

Few studies available in the literature address the classification of images of home environments using computer vision. The demand for such a task focused on real estate business applications has not yet been studied, and the requirements differ from those existing in applications such as robotics for which the problem has already been previously explored. In the present work, a classification system for furnished or unfurnished domestic environments is proposed. A convolutional neural network is used to classify images as belonging to the façade, kitchen, bathroom, bedroom, living room or external area environments. The results obtained by evaluating different architectures and using real images of the application domain show that the proposed approach allows to achieve accuracy of approximately 97%, being even superior to previous work focused on other application domains