Fall detection of elder people on thermal imaging using angular normalization, PCA, and weighted K-NN

In this paper we design an algorithm capable to classifying thermal images of people lying down and stand, to be applying it to a fall detection system. An automatic rotation, translation and size normalization algorithm was designed, to be applied of thermal image database, with the purpose of obta...

ver descrição completa

Detalhes bibliográficos
Autores: Ayala Raggi, Salvador Eugenio, Roa Escalante, Jesus Manuel, Barreto Flores, Aldrin, Portillo Robledo, José Francisco, Soid Raggi, Lourdes Gabriela, Bautista López, Verónica Edith
Formato: artículo
Estado:Versión publicada
Fecha de publicación:2022
País:México
Recursos:UNIVERSIDAD AUTÓNOMA DEL ESTADO DE HIDALGO
Repositorio:PÄDI Boletín Científico de Ciencias Básicas e Ingeniería del ICBI
Idioma:español
OAI Identifier:oai:repository.uaeh.edu.mx:article/9344
Acesso em linha:https://repository.uaeh.edu.mx/revistas/index.php/icbi/article/view/9344
Access Level:acceso abierto
Palavra-chave:Pattern recognition
PCA
Weighted K-NN
Image registration
Fall detection
Reconocimiento de patrones
K-NN ponderado
Registro de imágenes
Detección de caídas
Descrição
Resumo:In this paper we design an algorithm capable to classifying thermal images of people lying down and stand, to be applying it to a fall detection system. An automatic rotation, translation and size normalization algorithm was designed, to be applied of thermal image database, with the purpose of obtaining a new set of aligned images to performing a dimensionality reduction using PCA. Sequences of 100 frames were used, and both falling and non-falling sequences were produced. Applying the weighted K-NN classifier to identify the class of each frame, a probability vector of the lying class with 100 positions was obtained. These new vectors were used as training examples for a new K-NN classifier, which contains examples of falling and non-falling probability vectors. By applying cross-validation, the system is capable of recognizing falls with 91 % accuracy.