The MEP tracking vision system manufactured by FUJITSU is designed for real time tracking of multiple objects in the frames of a NTSC video stream. It consists of two VME-bus cards, a video module with frame grabber and overlay memory for displaying graphics and textual information, and a tracking module which can track over 100 templates at video frame rate (30Hz for NTSC). A MC68040/25MHz processor card running VxWorks executes the application program and controls the vision system.
Objects are tracked using template correlation. There are two sizes for templates available, 8x8 or 16x16 pixels, but both sizes can be magnified by a factor between 1 and 4 in the X- and Y-direction independently. The frames of the video stream are digitised by the video module and stored in dedicated video RAM which can also be accessed by the tracking module. Next, the templates are matched within specified search windows of the image. The comparison itself of the template and an area of the digitised frame is done by a cross correlation using the mean absolute error method. The distortion, which is the sum of the absolute greyscale differences of corresponding pixels in the template and the video frame, indicates how similar the two images are. Low distortions indicate a good match and high distortions appear when the images are very different.
To track the template of an object it is necessary to calculate the distortion not only at one point in the image but at a number of points within the search area. To track the motion of an object the tracking module finds the position in the search window where the template matches with the lowest distortion. By moving the search window along according to the tracking results objects can be tracked easily This method works perfectly for objects that do not change their appearance or shade and that get occluded by other objects.