UAV Aerial Surveillance to Georeference Aedes-Aegypti Mosquito Breading Grounds

Abstract

The annual number of cases of Dengue, Zika and Chikungunya in Brazil is alarming. One way to minimize this problem is to identify and treat potential mosquito breeding sites, such as swimming pools, unexplored water tanks, etc. With a focus on creating a tool to help authorities locate such mosquito breeding sites, this Master's Dissertation proposes a specific YOLO model to detect swimming pools in a given neighborhood, from images captured by a four-engine. Thus, a system is proposed object detection system based on a convolutional neural network, YOLOv3, to detect pools in images collected by the four-engine aircraft when flying autonomously over such a neighborhood. A dataset, with high resolution aerial images, with the desired objects annotated, was created to be used as a training dataset for YOLO's inner layers, with 150 images collected from photography websites high resolution. The evaluation of the classifier thus obtained occurred in a database containing 72 satellite images with different resolutions, in three different image scales, for two different locations, collected from Google Maps. Other tests were carried out on images collected by a four-engine Bebop 2 flying over a neighborhood, in addition to videos, frame by frame. The result is that the classifier got correctly detect the object, which means identifying the sought object and marking its location by means of a bounding box. Thus, the perspective of using the proposed system to detect possible mosquito breeding sites is quite significant, justifying the development of the entire system, as described in this dissertation.