Towards a Better Trade-Off Between Sensor Accuracy and Comfort in Smart Clothing Design: A Machine Learning Approach

摘要

Smart clothing technique is to embed sensors as well as other electronic devices into clothing to collect signals from human body and environment and make intelligent reaction accordingly. In its design, one of the biggest challenges is the difficulty to address both comfort and sensor accuracy in real-world application. Usually, skin-tight sensors yield good signal accuracy but uncomfortable feeling, while comfortable sensors loosely contacted with the skin lead to poor accuracy. In this work, we propose the idea that employs statistical machine learning approach to enhance the sensor performance by integrating the information from the inaccurate non-contact sensors, so that higher accuracy can be expected without making people uncomfortable. We justify the idea using the task of detecting body temperature using the sensors not in direct contact with the skin. We develop several types of features from the temperature sensors and integrate them in a non-linear regression model. The experimental results indicate that the method can improve the performance over 30% against the method of simple average and linear regression, which justifies the feasibility and potential of machine learning approach used for a better trade-off between sensor accuracy and comfort in smart clothing design.