Kalman Filter and 3D Warping Based Image Prediction for Realtime Computer Generated Image Sequences

摘要

Realtime computer based image generation is a challenging task in interactive simulation environments. In this context the unavoidable latency of such systems leads to various problems and degrades the performance of the visual system. In order to overcome this problem an image prediction algorithm is proposed which is based on McMillan's 3D image warping algorithm and the prediction of the future viewpoint by means of Kalman filtering. In other words, two algorithmic concepts stemming from completely different areas of research are combined in a favourable manner. Besides the reduction of latency effects the proposed method also admits to increase the frame rate of a given CGI (computer generated images) system. The quality of the predicted image sequence is the most interesting aspect. Hence, subjective and objective test results obtained via different realistic image sequences are discussed.