A Deep Learning Strategy for Stride Detection

In this paper a practical and convenient stride phase detection method is developed by using a torso-mounted inertial measurement unit (IMU). The proposed methodology generates labels that describes stationary and moving phases of the foot by preprocessing the sensor data provided by a foot-mounted IMU. While it is only a thresholding problem to detect the stride duration of a foot with a foot-mounted IMU, it is not trivial to achieve the same goal with an alternative configuration where the IMU is located at upper body. A secondary torso-mounted IMU captures data simultaneously where the data is divided into two groups as training and test samples to be used in a deep learning structure. Subsequently, two different deep learning structures are trained by using the corresponding labels provided by the foot-mounted IMU and the torso-mounted data. Promising results have been obtained with the developed method that is able to identify stride duration according to the torso-mounted IMU data. To the best of our knowledge, there is no other study that formulates the stride phase detection with a non-foot IMU using deep learning framework.