In-Ear Accelerometer-Based Sensor for Gait Classification

For several years, the detection of gait has been popularly implemented using wearable sensors, especially in the sports and medical areas. They are unobtrusive devices which allow to monitor individuals without the need of any ambulatory technology. Despite the fact, the optimal location of the sensor remains uncertain and dependent on the type of measurement. Ear-worn sensors provide a tactical position, robust against movement, that might be significant for gait classification. The purpose of this paper is to demonstrate the accuracy and reliability of in-ear accelerometer sensor to perform gait classification, between the activities walking and running. The data was collected from fourteen participants using an in-ear sensor called `Cosinuss° One', which contains a three-dimensional accelerometer sensor. The main characteristics between these two activities were detected using 17 time domain features, as for instance the maximums and standard deviations of the 3-axes, and 3 different window sizes were evaluated: 3.75s, 2s and 1s. Support vector machine (SVM) and k -nearest neighbors (KNN) classifiers were implemented and later compared. The total number of features was reduced to 6 for SVM and 12 for KNN preserving the same results. An accuracy over 99% for both classifiers was achieved, outperforming most of the previous studies.