Research Article
Identification and Classification of Human Body Parts for Contactless Screening Systems: An Edge-AI Approach
@INPROCEEDINGS{10.1007/978-3-031-06371-8_7, author={Diogo Rocha and Pedro Rocha and Jorge Ribeiro and S\^{e}rgio Ivan Lopes}, title={Identification and Classification of Human Body Parts for Contactless Screening Systems: An Edge-AI Approach}, proceedings={Science and Technologies for Smart Cities. 7th EAI International Conference, SmartCity360°, Virtual Event, December 2-4, 2021, Proceedings}, proceedings_a={SMARTCITY}, year={2022}, month={6}, keywords={Edge-AI Identification Classification Validation Thermal RGB}, doi={10.1007/978-3-031-06371-8_7} }- Diogo Rocha
Pedro Rocha
Jorge Ribeiro
Sérgio Ivan Lopes
Year: 2022
Identification and Classification of Human Body Parts for Contactless Screening Systems: An Edge-AI Approach
SMARTCITY
Springer
DOI: 10.1007/978-3-031-06371-8_7
Abstract
Continuous monitoring of vital signs like body temperature and cardio-pulmonary rates can be critical in the early prediction and diagnosis of illnesses. Optical-based methods, i.e., RGB cameras and thermal imaging systems, have been used with relative success for performing contactless vital signs monitoring, which is of great value for pandemic scenarios, such as COVID-19. However, to increase the performance of such systems, the precise identification and classification of the human body parts under screening can help to increase accuracy, based on the prior identification of the Regions of Interest (RoIs) of the human body. Recently, in the field of Artificial Intelligence, Machine Learning and Deep Learning techniques have also gained popularity due to the power of Convolutional Neural Networks (CNNs) for object recognition and classification. The main focus of this work is to detect human body parts, in a specific position that is lying on a bed, through RGB and Thermal images. The proposed methodology focuses on the identification and classification of human body parts (head, torso, and arms) from both RGB and Thermal images using a CNN based on an open-source implementation. The method uses a supervised learning model that can run in edge devices, e.g. Raspberry Pi 4, and results have shown that, under normal operating conditions, an accuracy in the detection of the head of 98.97% (98.4% confidence) was achieved for RGB images and 96.70% (95.18% confidence) for thermal images. Moreover, the overall performance of the thermal model was lower when compared with the RGB model.
