Classificação e monitoramento de áreas urbanizadas usando técnicas de visão computacional

Abstract

In this paper we propose a computer vision system to classify permeable and impermeable areas of a bounded area for study including the Micro-basin of Segredo and adjacent micro-basins, located in the municipality of Campo Grande/MS, Brazil, in order to evaluate the increase in urban density between the years 2008 and 2016. The proposed system is based on the image segmentation method Simple Linear Iterative Clustering (SLIC) to partition an image into multiple segments and generate superpixels that differentiate the permeable and impermeable areas; and attribute extraction algorithms to describe the visual features such as color, gradient, texture, and shape. The performance of five supervised learning methods was evaluated for the task of permeable and impermeable areas recognition. The proposed approach achieved an accuracy of 94.6% using the Support Vector Machine (SVM) algorithm. In addition, the results showed an increase of 7.2% in the urban occupation rate of the study area between the analyzed years. The results indicate that the proposed approach can support specialists and managers in the monitoring of urban density and its environmental impact.

Keywords

• visão computacional
• adensamento urbano
• impermeabilidade
• impacto ambiental