Authors:
Nikolajs Bumanis
1
;
Gatis Vitols
2
;
Irina Arhipova
2
and
Inga Meirane
2
Affiliations:
1
Faculty of Information Technologies, Latvia University of Life Sciences and Technologies, 2 Liela str., Jelgava, Latvia
;
2
WeAreDots Ltd., Elizabetes str. 75, Riga, Latvia
Keyword(s):
Face Recognition, Deep Learning, Long-term Identification.
Abstract:
A children face automated identification raise additional challenges compared to an adult face automated identification. A long-term identification is used in the environment in which a person must be identified in longer time spans, such as months and years. A long-term identification is present for example in schools where children spend multiple years and, if automated face identification solution is implemented, it must be resilient to recognise face biometrical data in the span of typically up to 9 years. In this proposal, we discuss children face identification available solutions which use deep learning networks, introduce legal constraints that come with privacy of children and propose prototype for a long-term identification of children attendance in their classroom. The solution consists of a developed prototype that is architecturally separated into three layers. The layers encapsulate necessary local and remote hardware, software and interconnectivity solutions between th
ese entities. The protype is intended for implementation into a school’s class attendance management system, and should provide sufficient functionality for person’s identity management, object detection and person’s identification processes. The prototype’s processing is based on the model that incorporates the principles of multiple correct biometric pattern versions, providing possibility of a long-term identification. The model uses Single Shot MultiBox Detector for object detection and Siamese neural network for a person identification.
(More)