Helping to detect legal swimming pools with Deep Learning and Data Visualization

Abstract The municipalities have, as a primary responsibility, the administrative management of their territory. The use of geographic information systems (GIS) and orthophoto maps can help to handle this task. One of the great difficulties they face is related to the continuous and quick changes that the territory suffers, and whose inspection is challenging to support. An example of this is the maintenance of the swimming pool registration system, in order to validate or license them. Indeed, it requires a substantial manual intervention, yet. This paper describes a system prototype for helping on detecting swimming pools on aerial images. It explores and integrates artificial intelligence (AI), systems integration, GIS, and data visualization. The AI improves the detection of objects, and middleware supports the integration of the detection results with other municipality systems for georeferencing and private property data licensing data crossing. To the innovative interoperability services, visualization libraries were explored, and an advanced visualization and analysis system was constructed. A dataset of aerial images of swimming pools and correspondent classification metadata was created. The model was trained with several convolutional neural networks in order to obtain and compare the precision results. The more accurate model is described.