Open AccessPedestrian age estimation based on deep learning
Author(s) -
Nawal Younis Abdullah,
Mohammed Ghazal,
Najwan Waisi
Publication year - 2021
Publication title -
indonesian journal of electrical engineering and computer science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.241
H-Index - 17
eISSN - 2502-4760
pISSN - 2502-4752
DOI - 10.11591/ijeecs.v22.i3.pp1548-1555
Subject(s) - convolutional neural network , computer science , artificial intelligence , deep learning , pedestrian , pedestrian detection , convolution (computer science) , computer vision , focus (optics) , popularity , face (sociological concept) , artificial neural network , pattern recognition (psychology) , machine learning , transport engineering , engineering , psychology , social psychology , social science , physics , optics , sociology
The large-scale distribution of camera networks in the traffic area resulted in the increasing popularity of video surveillance systems. As pedestrian detection and tracking are the critical monitoring targets in traffic surveillance, many studies focus on pedestrian detection algorithms across cameras. This paper addressed the effect of using the age estimation based on deep convolution neural network (CNN) as a convenience for pedestrian monitoring who is crossing at intersections. Two popular deep convolutional neural networks (DCNNs) pre-trained models have been used in this work, which have recently achieved the best performance in facial features extraction tasks: VGG-Face and ResNet-50. We combined these two models to increase the efficiency of the proposed system. We ran our experiments to evaluate the system based on the VGGFace2 dataset consisting of 3.31 million face images. From the experimental results, we observed a gap in the detection performances between those age groups: children from (00-10) years and elderly with 55 years and more. Moreover, it noted that the proposed pedestrian age estimation model performance is high, also a good result can be obtained by using the model for new purpose.