Research Of 3-D Motion Object Tracking Location Based On Stereo Sequence Images

The research target of computer vision is making the computer have the capability of cognizing three dimensions environment by means of two dimensions image, one of the aspects is making the computer perceive the geometric shape, location, pose and motion of the objects in three dimensions environment. This dissertation generally review and introduce the research developments of the sequence image three dimensions movements analysis in home and abroad present status, Combining the computer vision theory and photogrammetry methods systematically study the key contents such as Movement segmentation, Video camera calibration, Image match, Feature corresponding, Calculation of objects (three dimensions )movement parameters, objects movement evaluation and so on. Give out all algorithms of above-mentioned key sections and relevant tests, Substantial theory and test base are founded in the further research and practicability of the system.Camera calibration is the process of determining the transformation relationship between three-dimensional spatial coordinate system and two-dimensional camera image coordinate system. The precision and the reliability of camera calibration directly influence 3D measure precision of visic. system. Aiming at the fact that interior and exterior orientation elements of camera are required in accomplishing movement analyses based on feature correspondence when using the common video CCD camera, While the common video CCD cameras has many disadvantages such as interior orientation are unstable, higher lens distortion of objective, lower image resolution, photoelectric conversion deformation, undefined focus and principal point, etc. This dissertation proposes a new method. The author describes the methods of real-time in-suit calibration of CCD camera in the process of movement analyses based on feature correspondence by use of the known feature lines (points) which have geometric relation in object side. Sequential DLT algorithm is introduced to get interior and exterior orientation elements and lens photoelectric deformation errors. Object side coordinates of calibrated point are calculated as while as camera calibration.Movement segmentation detection is the key problem of object movement analysis, the aim of detecting moving object is picking up the change area from background image in sequence image. The valid segmentation of movement area is very important for the succeed object tracking location, thus in the following processing procedure, only pixels of movement area in the image need consider, this make the data manipulation more pertinence, Moreover, this still can save vast memory space. However movement detection it still a hardship work for the dynamic change of background image such as weather, illumination, shadow and the other disturbs. On the basis of the research practice, this dissertation separately adopt the method of dynamic object extraction based on difference of two frames and the method of dynamic object extraction based on background image difference for the dynamic object location on the motionless background. Results of experiment indicate: the method of dynamic object extraction based on difference of two frames has characteristic of simple and powerful operation, and combining the proper method of removing noise (such as the fast median filter andmorphology), valid result can be obtained; its disadvantage is the detective location of object is not sufficient precision. The background image difference method given in dissertation fully considers correlative information of long sequence image frames. Comparing with ordinary background image difference method (storage background image in advance, then difference with objective image), Restoring the background image with proper points which picking out from the whole image sequence on the basis of statistical rule has obvious advantages. The above two methods do not need prior knowledge of objective and background, can locate speedily , be applicable to the real-time application ,Images matching and fea...