Motion tracking in video sequences using features and regions.

摘要

Motion analysis plays a key role in the understanding of video imagery, both as a precursor to further analysis and a method for identifying points of interest in long spans of video. The focus of this research is an analysis of various techniques of motion and feature tracing in video sequences. The first of these methods is SURF feature detection to associate sequential images and detect motion trends. The second method is watershed region detection on sequential video frames, followed by template matching to detect the likely motion of each image region over time. The third technique tested was processing video sequences as a three dimensional volume using an extension of watershed regions over time. These three methods were studied for application in detecting basic camera techniques such as panning and zooming, as well as object motion on a static background.