An Experiment Comparing Double Exponential Smoothing and Kalman Filter-Based Predictive Tracking Algorithms

We present an experiment comparing double exponential smoothing and Kalman filter-based predictive tracking algorithms with derivative free measurement models. Our results show that the double exponential smoothers run approximately 135 times faster with equivalent prediction performance. The paper briefly describes the algorithms used in the experiment and discusses the results. 1 Double Exponential Smoothing-Based Prediction Double exponential smoothing-based prediction (DESP) is a viable alternative to the more common Kalman filterbased predictors with derivative free motion models. DESP models a time series using a simple linear regression equation where the y-intercept and slope are varying slowly over time[1]. An unequal weighting is placed on these parameters that decays exponentially through time so newer observations get a higher weighting than older ones. The degree of exponential decay is determined by the parameter α ∈ [0, 1). To predict user position, we assume that at time t−1, we have the estimates ~ b0(t− 1) and ~ b1(t− 1) for ~ β0(t− 1) and ~ β1(t − 1) respectively. Note that each estimate is a vector representing the x, y, and z components of position. We also assume we have a new user position ~ pt at time t. To update the estimates of ~ β0(t− 1) and ~ β1(t− 1), we require two smoothing statistics defined by ~ Sp t = α~ pt + (1 − α) ~ Spt−1 (1)