Moving object detection using unstable camera for video surveillance systems

Abstract

This paper presents a robust moving object detection method by compensating for motion of an unstable camera. Assuming that global camera motion results in affine transform between two successive frames, local affine motions are separately estimated in multiple pre-specified regions for fast, robust estimation of the global motion. The global camera motion is then estimated by the least squares method using the pre-estimated multiple local affine motions. Given a current frame as the reference, the subsequent frame is registered to the current frame using the estimated global motion. The moving objects are finally detected using difference of Gaussian and non-parametric kernel density estimation from the set of registered three frames. Experimental results show that the proposed method can robustly detect moving objects in unstable imaging environment for intelligent surveillance systems using various types of cameras including pan-tilt-zoom (PTZ) and unmanned aerial vehicle (UAV) cameras.