UAV Aerial Surveillance to Georeference Aedes-Aegypti Mosquito Breading Grounds

Name: ANTHONY OLIVEIRA PINTO

Publication date: 31/01/2022
Advisor:

Name	Role
MARIO SARCINELLI FILHO	Advisor *
PATRICK MARQUES CIARELLI	Co-advisor *

Examining board:

Name	Role
JORGE LEONID ACHING SAMATELO	External Examiner *
LUCAS VAGO SANTANA	External Examiner *
MARIO SARCINELLI FILHO	Advisor *
PATRICK MARQUES CIARELLI	Co advisor *

Summary: The annual number of cases of Dengue, Zika, and Chikungunya in Brazil are
alarming. A way to minimize this problem is to identify and treat possible mosquito breeding grounds, such as swimming pools, untapped water tanks, etc. Focusing on creating a tool to help authorities in finding such breeding grounds, as a contribution to the fight against these mosquitoes, this thesis proposes a YOLO custom model to detect swimming pools in a certain neighborhood, from images captured by a quadrotor. Thus, it is proposed an object detection system based on a convolutional neural network, the
YOLOv3 one, to detect swimming pools on images collected by the quadrotor when flying autonomously over such a neighborhood. A dataset was created, with high-resolution aerial bird’s eye view images, with the desired objects annotated, to be used as a training data set for the YOLO CNN inner layers, with 150 images collected from high resolution photography websites. The evaluation of the classifier thus obtained occurred on a database containing 72 satellite images with different resolutions, at three different image scales,
for two different locations, collected from Google maps. Other tests were performed over images collected by a Bebop 2 quadrotor flying over a neighborhood, besides videos, in a frame by frame basis. The result is that the classifier was able to correctly object detect, which means to identify the object searched for and to mark its localization through a bounding box. As a result, the perspective of using the proposed system to detect possible mosquitoes’ breeding grounds is quite meaningful, justifying the development of the whole
system, as described in this thesis.
Keywords: YOLO, CNN, object detection, Aerial surveillance, computer vision.

Access to document