In this paper we present a practical method for reducing the space when searching for point patterns in optically tracked virtual reality applications. The method uses a predictor to estimate the future position of a point. The key idea is to define a metric that determines the quality of the predictor, and use this metric to construct a 3D region around the predicted position of each point. The size and shape of the 3D region is based on the kinematic properties of the predicted point. The 3D region is projected into 2D to obtain the required search window.The contribution of the paper is that the search space can be reduced substantially by making use of adaptive window shapes and sizes.