Constrained Particle Filter for Improving Kinect Based Measurements

Microsoft Kinect has been gaining popularity in home-based rehabilitation solution due to its affordability and ease of use. It is used as a marker less human skeleton tracking device. However, apart from the fact that the skeleton data are contaminated with high frequency noise, the major drawback lies in the inability to retain the antropometric properties, like the body segments' length, which varies with time during the tracking. In this paper, a particle filter based approach has been proposed to track the human skeleton data in the presence of high frequency noise and multi-objective genetic algorithm is employed to reduce the bone length variations. In our approach multiple segments in skeleton are filtered simultaneously and segments' lengths are preserved by considering their interconnection unlike other methods in available literature. The proposed algorithm has achieved MAPE of 3.44% in maintaining the body segment length close to the ground truth and outperforms state-of-the-art methods.