Detecting People from Beach Images

Abstract

To avoid risks inherent to aquatic environments,such as drownings and shark attacks, some beach areas must bemonitored continuously. If needed, a rescue team has to respondas quickly as possible. This project puts forward a proposal ofan algorithm for people detection as part of a system that willautomatically monitor people in the sea and at the beach areas inorder to help lifeguards prevent these risks. The major challengesto solving this problem are: variable brightness on cloudy days,the position of the sun at different times of the day, the difficultyin segmenting an image, seeing partially submerged people, andthe position of the camera. For person detection, a commonpractice found in the literature is to use image descriptors inconjunction with a fast and accurate classifier for a real-timesystem. This study examines a data set of beach images using thefollowing image descriptors and their pairwise combinations: Humoments, Zernike moments, Gabor filter, Histogram of OrientedGradients (HOG) and Local Binary Pattern (LBP). Furthermore,a dimensionality reduction technique (PCA) is used for featureselection. The detection rate is evaluated with the followingclassifiers: Support Vector Machine (SVM) with linear and radialkernels, and Random Forest. The experiments demonstrate thatthe SVM classifier with a radial kernel using the HOG and LBPdescriptors with PCA showed promising results, 90.31% accuracy being obtained.

Publication
In 2017 International Conference on Tools with Artificial Intelligencs
Date
Links