TCC - Bacharelado em Ciência da Computação (Sede)
URI permanente para esta coleçãohttps://arandu.ufrpe.br/handle/123456789/415
Navegar
1 resultados
Resultados da Pesquisa
Item Segmentação de banhistas utilizando algoritmos de agrupamento com seleção automática do número de grupos em regiões litorâneas(2019) Moura, Allan Alves de; Macário Filho, Valmir; http://lattes.cnpq.br/4346898674852080; http://lattes.cnpq.br/3319938637009294The increasing number of shark attacks has been frightening people that lives in coastal areas, making it impossible to bath in certain places. In an attack situation, most of the time a course of action to save the victim’s life is only taken after the incident already has occurred, which a lifeguard tries to help her. An auxiliary tool for lifeguards was thought in order to mitigate these events and allow the lifeguards to act before the incident happens, alerting the professional if someone tries to surpass a delimited zone. The first step to bring this auxiliary tool to life is the technique of image segmentation on beach photos in search for regions that share visual similarities in order to find people inside the sea. Therefore the objective of this work is to study and find a good image segmentation algorithm capable of automatically selecting the best number of groups without the parameter control necessity. The selected algorithm will be used to implement the first phase of the lifeguard auxiliary tool in search for image regions that represent bathers. Image pre-processing techniques like beach removal were evaluated, as well as characteristics vectors selection used to compare elements. The combination between algorithms and characteristics vectors were evaluated with and without beach removal. The analyzed algorithms were: hierachical aglomerative, hierarchival divisive, X-means, auto group segmentation and automatic colored image segmentation. All of them were applied to three different characteristics vectors composed by the color system RGB (red, green and blue), LAB and the combination of RGB + LAB. The most promising result, taking into account the visual analysis and the algorithm comportamental analysis, was obtained by the automatic color image segmentation with RGB+ LAB, composed characteristics vector, with value of 1.5245 extracted from Dunn’s index, using the beach removal as post-processing technique.