Enhancing Motion Tracking with Kalman Filter and Tobit Kalman Filter Techniques

Abstract

In this paper, two methods are proposed to analyse skeleton data recorded by the Kinect v2 sensor using the Kalman filter and the Tobit Kalman filter in order to minimize the noise of the acquisition device due to occlusions, self occlusions e.t.c. The skeleton data are three-dimensional spatial coordinates that record movements of an individual’s joints. The variance of the noise process is estimated using the likelihood function. In order to include into the model restrictive conditions based on the joints displacements per frame, we apply the Tobit Kalman Filter. Experiments on skeleton data show that the Tobit Kalman filter corrects the noise better than the standard Kalman filter.