Model of Computation

Abstract

Computer vision transcends a wide range of representations and forms of processing. Despite advances in many sub-areas of computer vision, there is no consensus on a unified approach to vision. However, one can define certain general characteristics of an Integrated Vision System (IVS) from computational perspective. For example, it is known that a “vision system” must be able to perform diverse sets of complex operations on a massive amount of data at high speeds. Motion sequences at moderate resolution (512×512 pixels) and typical frame rate (30 frames/sec) in color (3 bytes) involve more than 20 Mbytes of data per second. The amount of computation required for dynamic scene interpretation including labeling objects, surface reconstruction and motion analysis is difficult to estimate; however, for many applications computational power in the range of 10^12–14 instructions per second is required [41]. Not only are the raw processing needs tremendous, but varying the type of processing capabilities (such as number crunching, symbol manipulation, and data processing) are required.