Motion vectors based motion tracking

Abstract

Computer vision methodologies have become widely applicable over the last few years due to the extensive areas of research they encapsulate, from motion estimation techniques to object detection, tracking, and recognition. Over recent years, the gaming industry has surged forward by employing such methods. These have resulted in an overall better gaming experience through custom-made natural user interfaces. The following study is based on the design and implementation of a partial MPEG-2 software decoder, allowing the extraction and calculation of motion vectors as it parses though a video stream. In order to do so, the structure of the picture was considered, starting from slices and proceeding to the macroblock level. The considered video clip was stripped from any audio content in the aim of eliminating unnecessary processing overheads. The motion vectors extracted from a short tennis sequence were then displayed in a C++ and Win32 GDI environment using Visual Studio 2008 C++ Express Edition. This led to the second stage of this study. Sets of coordinates corresponding to areas with the highest degree of motion were calculated in an attempt to localize the racket. A resynchronization mechanism was used to determine whether the coordinates give a reliable indication of the racket location, in which case the corresponding calculated speed is output. The bounding box concept was then applied to a partitioned picture. The output given for each partition represents the centre of a region of motion exceeding a threshold value. The threshold was set low, leading to the elimination of irrelevant and small movements contained within the picture. An evaluation of the results then follows. The system proved to be successful at locating regions with large motion vectors, making this system an ideal basis for tracking-related systems.