Residual Network of Residual Network: A New Deep Learning Modality to Improve Human Activity Recognition by Using Smart Sensors Exposed to Unwanted Shocks

Document Type: Original Article


1 Faculty of Electrical, Biomedical and Mechatronics Engineering, Qazvin Branch, Islamic Azad University, Qazvin, Iran,

2 Associate Professor of Biomedical Engineering, Iranian Research Organization for Science and Technology, Tehran, Iran


Background and Objective: Recently, smartphones have been vastly utilized in monitoring the daily activities of people to check their health. The main challenge in this procedure is to distinguish similar activities based on signals recorded by using sensors mounted on smartphones and smartwatches.
Although deep learning approaches have better addressed the above challenge than alternative methods, their performance may be severely degraded, especially when the mounted sensors receive disturbed signals due to smartphones and smartwatches not being in a fixed position.
Methods: In this article, a new deep learning structure is introduced to recognize challenging human activities by using smartphones and smartwatches, even when the recorded signals are noisy due to the sensors being unstable. In the proposed structure, the residual network of residual network (i.e. ROR) is engaged as a new concept inside the deep learning architecture, which provides greater stability against either disturbed or noisy signals.
Results: The performance of the proposed method is evaluated on recorded signals from smartphones and smartwatches and compared with the state of art techniques containing deep learning and classic (non-deep) schemes. The obtained results show that the proposed method may improve the recognition parameters at least 1.79 percent against deep alternatives in distinguishing challenging activities (i.e. downstairs and upstairs). These superiorities reach at least 32.86 percent for classic methods, which are applied on the same data.
Conclusions: The effectiveness of the architecture in recognizing either challenging or non-challenging activities in the presence of unwanted cell phone shocks demonstrates its potential to be used as a mobile application for human activity recognition.