Hardware acceleration for tracking by computing low-order geometric moments

With the growing number of video content analysis applications, efficient implementation has become increasingly important. Video-object tracking using image moments is an important subtask in video-content analysis content algorithms. In this paper, we will present a method of accelerating the computation of geometrical moments and the resulting moment engine to a throughput of 130-fps. Furthermore, the effect of the accelerator on the performance of two object-tracking applications in a multi-processor platform will be evaluated. The conclusion is that the moment engine is relatively inexpensive in terms of required gate area, but its integration efficiency into a domain-specific multi-processor platform remains to be further analyzed.