Detecting People from Beach Images

To avoid risks inherent to aquatic environments, such as drownings and shark attacks, some beach areas must be monitored continuously. If needed, a rescue team has to respond as quickly as possible. This project puts forward a proposal of an algorithm for people detection as part of a system that wi...

Full description

Saved in:

Bibliographic Details
Published in:	2017 IEEE 29th International Conference on Tools with Artificial Intelligence (ICTAI) pp. 636 - 643
Main Authors:	Luna da Silva, Ricardo, Chevtchenko, Sergio, Alves de Moura, Allan, Rolim Cordeiro, Filipe, Macario, Valmir
Format:	Conference Proceeding
Language:	English
Published:	IEEE 01-11-2017
Subjects:	Feature extraction Forestry Histograms image classification image descriptors Image segmentation Kernel Monitoring people detection random forest Support vector machines svm
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	To avoid risks inherent to aquatic environments, such as drownings and shark attacks, some beach areas must be monitored continuously. If needed, a rescue team has to respond as quickly as possible. This project puts forward a proposal of an algorithm for people detection as part of a system that will automatically monitor people in the sea and at the beach areas in order to help lifeguards prevent these risks. The major challenges to solving this problem are: variable brightness on cloudy days, the position of the sun at different times of the day, the difficulty in segmenting an image, seeing partially submerged people, and the position of the camera. For person detection, a common practice found in the literature is to use image descriptors in conjunction with a fast and accurate classifier for a real-time system. This study examines a data set of beach images using the following image descriptors and their pairwise combinations: Hu moments, Zernike moments, Gabor filter, Histogram of Oriented Gradients (HOG) and Local Binary Pattern (LBP). Furthermore, a dimensionality reduction technique (PCA) is used for feature selection. The detection rate is evaluated with the following classifiers: Support Vector Machine (SVM) with linear and radial kernels, and Random Forest. The experiments demonstrate that the SVM classifier with a radial kernel using the HOG and LBP descriptors with PCA showed promising results, 90.31% accuracy being obtained.
ISSN:	2375-0197
DOI:	10.1109/ICTAI.2017.00102