A Novel Zero-velocity Detector for Pedestrian Inertial Navigation based on Deep Learning

Abstract

Due to the poor accuracy of the MIMU (Miniature Inertial Measurement Unit), ZUPT (zero-velocity update algorithm) is essential for foot-mounted inertial pedestrian navigation. Zero-velocity detection is the premise of ZUPT. Although conventional threshold-based zero-velocity detectors have achieved nice results in detection, it needs to tune thresholds and is not available in all cases. Modeling the precise mathematical relationship between thresholds and foot motion is hard, because lots of factors can affect the human foot motion. The zero-velocity detection is actually the binary classification which can realized by machine learning. We designed a novel zero-velocity detector based on deep learning. The convolutional neural network is used to train the detector based on the preprocessed inertial data. This detector does not need a motion classifier and have a better generality. Experiments are conducted and results show that this novel zero-velocity detector can perform better in combined motion than the threshold-based zero-velocity detector.