University of Houston

Change detection from video sequences is critical in many applications related to video surveillance, remote sensing, and medical diagnosis. The main goal of video registration algorithms for change detection is to identify the set of pixels that are significantly different between spatially aligned images that are temporally separated. This problem is extremely challenging due to a variety of factors, including changes in the illumination over time, appearance or disappearance of objects, and temporal misalignment of video sequences. Further, the exact trajectory of the mobile video acquisition platform is not controlled, which results in a change in scale of the observed scene along with rotation and translation changes between image pairs. Finally, due to the varying field-of-view, the imaging geometry cannot be modeled by affine constraints. Hence, the image sequences exhibit perspective and parallax. Over the years, many standard image processing techniques have been leveraged to realize a solution to the problem of change detection. Each potential approach attempts to exploit properties of the image, the application domain, or a combination. The relevance of the kind of changes to be detected is application specific, but the underlying algorithms need to detect all changes as the first step, which can later be post-processed to discriminate between relevant and unimportant changes. In this work, we explore more complex imaging models for solving the video registration task and propose a framework that accomplishes spatiotemporal registration and change detection in six steps:(1) temporal alignment of the two video sequences; (2) disparity estimation of scenes; (3) refinement of the individual disparity estimates using sequential stereo images; (4) dominant plane segmentation using texture and disparity estimates; (5) estimation of spatial transform for the dominant plane; and (6) detection of textural anomalies to delineate areas of change. In the last decade, video surveillance has become one of the most significant and active research areas due to increasing security concerns. There is a definite need for a change detector that first automatically correlates and compares two video sequences taken of the same scene from different viewing angles at different instances and then displays the changes and their locations. We will employ our framework to detect the regions of change between two different video sequences recorded by a mobile uncalibrated stereo video platform of the same dynamic outdoor environment where threatening objects may be placed.