Step Length Estimation Using Sensor Fusion

Abstract

One of the main challenges of navigation systems is the inability of orientation and insufficient localization accuracy in indoor spaces. There are situations where navigation is required to function indoors with high accuracy. One such example is the task of safely guiding visually impaired people from one place to another indoors. In this study, to increase localization performance indoors, a novel method was proposed that estimates the step length of the visually impaired person using machine learning models. Thereby, once the initial position of the person is known, it is possible to predict their new position by measuring the length of their steps. The step length estimation system was trained using the data from three separate devices; capacitive bend sensors, a smart phone, and WeWALK, a smartcane developed to assist visually impaired people. Out of the various machine learning models used, the best result obtained using the K Nearest Neighbor model, with a score of $0.945 R^{2}$. These results support that indoor navigation will be possible through step length estimation.