Overview

Efforts must be directed towards matching the system architecture both to the image data structure and to the image processing algorithm.

Cantoni and Levialdi [52 ]

Developing an appropriate motion estimation algorithm is only the first part of creating a practical intelligent motion sensor for use in autonomous navigation. Once the algorithm is verified and its operational parameters defined, it must be implemented so that those parameters are met. The algorithm developed in this book is relatively simple, so implementation using modern, general purpose processing hardware such as the humble PC is feasible. Use of a PC would also simplify development and testing thanks to the large and mature set of development tools available. However, a PC based solution has a number of drawbacks when building an intelligent sensor. Using a PC certainly does not lead to a compact solution. Indeed, it leads to a higher parts count since one cannot directly interface to the processor: all I/O devices must be connected to the PC via special purpose add-on cards. A PC also consumes a large amount of power. Another drawback is that a standard PC gives no direct way of utilising the parallelism available in the algorithm. Special purpose image processing hardware may allow the use of parallelism, however, the other drawbacks remain.

Building a CPU or DSP based system (or using one of the many available development platforms) allows more control over system design. Such systems are generally more compact than a fully-fledged PC since they...