Tracking Filter Framework

Virtual reality is a computer simulation wherein a system tracks the user’s pose and replaces sensory feedback for one or more senses to place the user into a Virtual environment (VE). In order to immerse users into this environment, the system needs to be accurate and fast in capturing their actions and feed their representations back to the senses. Measurement and pose estimation errors are a common problem for such systems but can be mitigated through the use of filter algorithms. This work documents the design and implementation of a Tracking Filter Framework (TFF), and evaluates its ability to reduce tracking errors and enhance the user experience. The TFF applies filter algorithms to tracking data and provides the result as an output. The Lighthouse Tracking System (LHTS) with Valve’s Index is used as tracking data source, it supports six Degrees Of Freedom (DOF) to track the user’s pose and uses optical and inertial tracking. The experimental library libsurvive is used to access the inertial tracking data. The Double Exponential Smoothed Prediction (DESP), a double exponential filter, and the Error-state Kalman Filter (ESKF), an error-state Kalman filter capable of fusing optical and inertial data, are introduced. The user acceptance of the framework is evaluated by conducting a user study within a VE, using a within-subject design. The results show that the provided tracking data by libsurvive with disabled optimizations turned out to be too noisy and unstable for the introduced filters. The filters cannot compensate for the occurring tracking errors to the degree that would have been necessary for a Virtual Reality Application (VRA). The DESP causes a significant delay when trying to filter noisy tracking data, which is not acceptable for a VRA. The ESKF provides a significant improvement with simulated tracking data. However, it falls short with libsurvive tracking data because of its noisy and unstable nature, as the results of the user study show.