Low-Latency Filtering of Kinect Skeleton Data for Video Game Control
Matthew Edwards, Richard Green · 2014
The Kinect sensor was originally designed for use as a controller for video games, however the noise and latency in the output of its skeleton tracking system negatively impact user experience. This paper compares four different filter-based approaches to mitigating that noise and latency in a simple video game: the Kinect SDK's built-in Holt double exponential smoothing filter, an averaging filter, a Kalman filter with a constant-value model, and a Kalman filter with a Wiener Process Acceleration (WPA) model. The averaging filter and constant-value Kalman filter both have good smoothing performance but introduce relatively large amounts of latency, while the others have good performance in both measures. The WPA Kalman filter was found to have the best overall performance.