Robust and accurate zero-velocity detection can improve the performance of zero-velocity-aided foot-mounted inertial navigation system. To ensure the accuracy of zero-velocity detection, we propose a novel detector based on contrastive learning. This detector roughly eliminates the inertial data that must not be the zero-velocity event in advance, to reduce the computation cost. Then the detector uses the remaining inertial data to detect the zero-velocity event via a trained contrastive neural network. The contrastive neural network uses the triplet network and is trained by comparing with the anchor data which consists of the known static inertial data from the period of initial alignment. The classifier will finally determine whether the output of the triplet network is the zero-velocity event. Two experiments were conducted to evaluate this novel detector, showing that it can adaptively and accurately detect the zero-velocity event. The horizontal position errors of the two experiments are respectively 1.33m over a 953m outdoor path with walking and 3.74m over a 1143m indoor/outdoor path with combined motion of low dynamic and high dynamic.

The detection of zero-velocity states is the vital prerequisite for zero-velocity update in the foot-mounted inertial pedestrian navigation system. The previous zero-velocity detector determines zero-velocity states by comparing measured inertial data with a calibrated threshold. The calibration of the threshold is inconvenient for this kind of the zero-velocity detector because the threshold is variable corresponding to different people and locomotion. The best threshold needs to be tuned corresponding to different situations. In essence, the detection of zero-velocity states is a binary classification problem. As the success of deep learning in in image classification and speech recognition, it is possible to design an adaptive zero-velocity detector based on it. A Siamese network is designed to learn the metric of distinguish zero-velocity states. This method can adaptively get the most likely correct results without threshold tuning. Experiments are conducted and results show that the matching degree is about 96.31% and the navigation accuracy can reach within 4m in 20min.