The use of Kalman filters in human gait analysis using depth sensors

Four variants of the Kalman filter, based on different mathematical models of movement, have been used for denoising measurement data representative of human gait. A modification of the Kalman filter, aimed at detecting and suppressing errors of a kind typically corrupting data acquired using depth sensors, has been proposed and studied. The considered variants of the Kalman filter have been tested using real-world and semi-synthetic data. The experimental results indicate that the Kalman filter, combined with a suitable model of movement and the proposed modification, allows for effectively denoising data from depth sensors and may yield better results than the commonly used Butterworth filter.