Analyzing gait symmetry with automatically synchronized wearable sensors in daily life

Abstract Gait deviations such as asymmetry are one of the characteristic symptoms of motor dysfunctions that contribute to the risk of falls. Our objective is to measure gait abnormalities such as asymmetry of the lower limbs in order to evaluate the diagnosis more objectively. For the measurement we use inertial measurement unit (IMU) sensors and force sensors, which are integrated in wristbands and insoles. To extend the battery life of wearable devices, we only save data of the activity gait within the wearables. Therefore we perform activity recognition with a smartphone. Using convolutional neural network (CNN) we achieved an accuracy of 94.7 % of the activity gait recognition. Before recording we synchronize the wearable sensors and reach a maximum latencies of 3 ms. Before the analysis of the symmetry we detect the strides by using a CNN with an accuracy of 98.8 %. For the symmetry evaluation we used dynamic time warping (DTW). The DTW enables us to calculate symmetry of the complete time series of human gait.