An Evaluation of Background Subtraction Algorithms on Fused Infrared-Visible Video Streams

Abstract

The detection of motion is an essential preprocessing step in many vision based systems. While showing good performance in the visible or the infrared spectrum, some of the state-of-the-art background subtraction methods are quite sensitive to a change in the spectral range. In this paper, the robustness of various background subtraction algorithms is not only compared between visible and infrared video streams, but in addition to the robustness that can be achieved by fusing visible and infrared video streams. Thereby, we show the effects of several fusion methods on a large set of background subtraction algorithms. By analyzing quantitative results, we identify approaches which can benefit from fused sensor signals. Towards this end, we further analyze the effectiveness of 14 fusion strategies. The evaluation is done on the public available OSU Color-Thermal Database reflecting a typical outdoor surveillance scenario.