Autonomous Detection of Mosquito-Breeding Habitats Using an Unmanned Aerial Vehicle

Every year, thousands of people die from diseases such as dengue fever, chikungunya, and zika. This public health problem is more evident in developing countries. All these illnesses have the same source in common: a mosquito scientifically known as Aedes Aegypti. In cities, the majority of mosquito-breeding habitats are man-made: bottles, tires, barrels, pots or any stagnant water. This paper proposes an innovative system to aid in determining mosquito-breeding habitats location by employing computer vision tools on aerial images. Tires and regions with stagnant water were selected as objects of study. For this, a dataset was created containing video sequences and telemetry data from an unmanned aerial vehicle (UAV) and the respective manual annotation in different scenarios. The features extracted from the videos (through HSV color space, histograms, and edge detection) were used to train a random forest classifier, resulting in an accuracy higher than 99% in the test set. The system is also capable of automatically determining the GPS coordinates of the possible mosquito breeding location with similar precision to commercial GPS equipment.