A comparative study of applying low-latency smoothing filters in a multi-kinect virtual play environment

Abstract

The Skeleton Tracking System in Kinect is known for being noisy and unstable, hence, in practice, a noise reduction filter or smoothing filter needs to be employed before consuming the data in order to obtain smooth joint position data over time. In this paper, we present a comparative study on applying four different smoothing filters (Simple Moving Average Smoothing, Savitzky–Golay filter, Exponential filter, and Double Exponential filter) in “Alone Together” (Tang et al. 2015), a virtual play environment augmented with multiple sets of Kinects. Overall, among the four filters, the Exponential Smoothing Filter yields the best results in the game. The comparative study only provides quantitative observations on the four smoothing filters, the qualitative examination in terms of player satisfaction remains unclear, which is one of our immediate future research paths in this direction.