Robust Data-Driven Zero-Velocity Detection for Foot-Mounted Inertial Navigation

Abstract

We present two novel techniques for detecting zerovelocity events to improve foot-mounted inertial navigation. Our first technique augments a classical zero-velocity detector by incorporating a motion classifier that adaptively updates the detector's threshold parameter. Our second technique uses a long short-term memory (LSTM) recurrent neural network to classify zero-velocity events from raw inertial data, in contrast to the majority of zero-velocity detection methods that rely on basic statistical hypothesis testing. We demonstrate that both of our proposed detectors achieve higher accuracies than existing detectors for trajectories including walking, running, and stairclimbing motions. Additionally, we present a straightforward data augmentation method that is able to extend the LSTMbased model to different inertial sensors without the need to collect new training data.

Machine Learning-Based Zero-Velocity Detection for Inertial Pedestrian Navigation

“…These data should be labeled to train the model in a supervised way. Existing research work performs the data labelling with extra sensors (e.g., pressure sensor [20]), manual annotation according to certain criteria [21], or with the help of a high accuracy motion capture system [13]. Then the training and validation procedures is done either by machine learning or deep learning approach.…”

“…Two foot-mounted inertial navigation benchmarking solutions are used to assess the performance of our approach. The first one is the recently published AI based ZUPT detection method in [13], labeled LSTM utiasSTARS in the rest of the paper. The second solution is the Extended Kalman Filter based software suite [14], labeled PERSY.…”

“…To run LSTM utiasSTARS on our evaluation dataset, we used the model provided by [13] in the github. Since all training parameters are not provided (e.g.…”

“…The most advanced approaches rely on deep leaning techniques that are sometimes combined with parallel motion classification methods. They provide more robust ZUPT detection performances [13] but rise questions about the need for large training databases and even heavy computation costs, preventing real time applications.…”

FootSLAM meets Adaptive Thresholding

“…A common approach is to first estimate or classify the speed or motion mode of the user. Based on the result, the detector selects a threshold value that has been optimized, using ground truth data, for that specific speed or motion class [4]- [9]. However, other calibration methods have also been proposed.…”