Pedestrian Stride-Length Estimation Based on Bidirectional LSTM Network

Stride-length estimation is an important part of Pedestrian Dead Reckoning (PDR). In view of the problem that traditional stride-length estimation model has too large estimation errors in complex environments and special gaits, a pedestrian stride-length estimation algorithm based on Bidirectional LSTM Network is proposed to realize accurate estimation of stride-length in normal walking, fast walking, slow walking, running and jumping gait. The algorithm takes raw inertial data of accelerometer and gyroscope as the input and the stride-length as output, which can effectively process the time-dependent inertial data within a gait cycle, so as to extract the relevant features of pedestrian stride-length. The effectiveness of the algorithm is verified by collecting actual data from the built-in inertial sensor of the smartphone. The average stride-length estimation relative error rate is 2.80%, and the average distance estimation error rate is 0.95%, which shows that a good estimation accuracy has been achieved.