Movement prediction in rowing using a Dynamic Time Warping based stroke detection

Abstract

In professional rowing competitions, sensor data is transmitted from an on-board sensor unit on the boat to an external computer system. This system calculates the current position of each boat in real-time. However, incomplete localizations occur as a result of radio transmission outages. This paper introduces an algorithm to overcome transmission outages by predicting the rowing movement. The prediction algorithm is based on accelerometer and GPS data that is provided by the on-board unit before an outage occurs. It uses Subsequence Dynamic Time Warping (subDTW) to detect the rowing strokes in the acceleration signal. Knowing the previous strokes, the system predicts the upcoming strokes, as the rowing motion follows a periodic pattern. Thereby, the GPS measured velocity can be extrapolated and the position is predicted. A further outcome of the subDTW stroke detection is an accurate determination of the rowing stroke rate. In our experiment, we evaluate the rowing stroke detection and stroke rate determination based on subDTW as well as the prediction algorithm for simulated outages of professional race data. It shows a subDTW stroke signal detection of 100% after the start phase of the race. The prediction in case of a sensor outage of 5 seconds leads to a correlation between the predicted velocity and the actual velocity of 0.96 and a resulting position error (RMSE) of 0.3 m.