Stride-Length Estimation for Indoor Navigation

Parkinson’s disease can easily lead to falls for the elderly. The pedestrian dead reckoning system can identify the behavior and location of the elderly and promptly warn to the caregivers [1]. However, the accuracy of stride-length estimation is crucial to achieving precise positions in personal indoor positioning systems [2]. Currently, stride estimation methods mainly include constant-based stride estimation models, nonlinear stride estimation models, and artificial intelligence-based stride estimation models, etc. However, it suffers from some issues such as low estimation accuracy and limited applicability. Martinelli et al. [3] proposed a stride regression model based on contextual information, which has high accuracy, but it cannot reflect the error caused by complicated environmental factors. Wang et al. [4] firstly used the LSTM for time series feature extraction, and then used the extracted features to estimate the pedestrian’s stride length through the neural network model. Hannink et al. [5] proposed a stride estimation method based on deep convolutional neural networks, which had high accuracy. However, this method relies on specific shoe-mounted inertial sensors and is unsuitable for smartphones.