Detecção de pessoas em um ambiente industrial utilizando imagens de profundidade e classificadores profundos

Eduardo Henrique Arnold, Danilo Silva

DOI: 10.14209/sbrt.2017.132

Evento: XXXV Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2017)

Keywords: Human detection depth images deep learning convolutional networks machine learning computer vision

Abstract

This paper describes the development of an indus- trial safety system that requires automatic human detection. Two solutions based on top-view depth images are presented. The first one is based on traditional learning techniques using feature extraction and a Support Vector Machine classifier. The second solution uses deep learning methods for classification. The performance analysis of both solutions revealed that the deep learning methods outperform traditional learning techniques on this task, at the cost of requiring a larger training set and increased computational cost.

Download