Deep-Learning-Based Biometric Methods for Security in Smart Environments

This paper considers, proposes, and describes ways and possibilities to implement various biometric methods based on deep learning in different types of smart environments. The main objective is to investigate and propose how these methods can be applied in smart environments such as office buildings, collective residential buildings, and houses, primarily to enhance security. To this end, three biometric methods are analyzed and proposed for use: gait, face, and palmprint recognition. Accordingly, this paper describes possible applications of gait, face, and palmprint recognition for person identification in smart environments, specifically in an office building and a house. Both gait and face recognition can be used without interaction with the person being identified. Palmprint recognition, however, requires interaction, as the person must place their palm on a designated device. These three methods were developed using deep learning models. The TensorFlow platform and Keras are used for model development. The architecture of the models and related settings are described. The training and validation processes were conducted using four datasets that were publicly available or available on request: a gait dataset, a face dataset, and two palmprint datasets. The results primarily related to validation accuracy are presented in the paper. In addition to validation accuracy, results for precision, recall, and F1-score are also reported. Furthermore, the reasons for choosing these three methods, along with their advantages and disadvantages, are discussed.