Graffiti and government in smart cities: a Deep Learning approach applied to Medellín City, Colombia

Abstract

Graffiti is an element of graphic expression that manifests different states of the human being. However, for many governments worldwide, it has been an element of discord between them and the communities that express themselves through graffitis. This article proposes identifying graffiti and concentration zones through Computer Vision and object detection and localization to support public policy management in smart cities. ASUM-DM methodology is used to achieve the aim. Initially, the current problems faced by municipal governments in the management of public graffiti policy are identified. Then available datasets of images from Google Street View (GSV) and other acquired datasets are identified for the case study carried out in the city of Medellín (Colombia) and border municipalities. A training dataset of 1,395 images and a production dataset of 71,100 panoramas is placed on strictly using the experimental method of the division of training data, validation, and a production sample, to make a correct estimation of the generalization error. As a result of the training process, we obtained an Average Precision of 69,14%, which presented a high precision Tag of 89.23%, and low precision of 59.13% in Mural. Finally, it is possible to build heat maps of graffiti concentration areas that could guide rulers to create or improve public policies related to graffiti expression.