**2.2.2 Single-sensor device**

This architecture reduces cost by placing a colour filter array (CFA), which is a mosaic of colour filters, on top of the conventional single CCD/CMOS image sensor to capture all

Machine Vision Measurement Technology and Agricultural Applications 135

simulations or algorithmic analysis. During software development, careful tracking of central processing unit utilization is needed to focus on those code units that are slow or that have response times that are inadequate. Unfortunately, cache and direct memory access, which are intended to improve average real-time performance, destroy determinism and thus make prediction of deadlines troublesome, if not impossible. But, schedulability analysis is usually the subject of traditional texts on real-time systems

An understanding of the hardware support for colour imaging graphics is fundamental to the analysis of real-time performance of the system. Some specialized hardware for realtime imaging applications involves high-performance computers with structural support for complex instruction sets and imaging coprocessors. Inexpensive pixel processors are also available, and scalable structures are increasingly being used for real-time imaging applications. But building systems with highly specialized processors is not always easy, particularly because of poor tool support. Therefore, many commonly deployed colour imaging systems use consumer-grade personal computers. There are many architectural issues relating to real-time performance, such as internal/ external memory bus width, memory access times, speed of secondary storage devices, display hardware issues, and colour representation and storage, to name a few. Collectively, these design issues involve three trade-off problems — schedulability versus functionality, performance versus resolution, and performance versus storage requirements. Real-time design of imaging systems, then, involves making the necessary decisions that trade one quality for another, for example, speed versus resolution. For example, one of the main performance challenges in designing real-time image processing systems is the high computational cost of image manipulation. A common deadline found in many processing systems involves screen processing and update that must be completed at least 30 times per second for the human

Because this processing may involve more than a million pixels, with each colour pixel needing one or two words of storage, the computational load can be staggering. For the purposes of meeting this deadline, then, the real-time systems engineer can choose to forgo display resolution, or algorithmic accuracy. Finding improved algorithms or better hardware will also help meet the deadlines without sacrifice — if the algorithms and hardware behaviorare bounded, which, as mentioned, is not always the case. In this section, however, we confine our discussion to two important hardware issues (Lukac & Plataniotis, 2007).

Our interest is in the appropriate representation of colour in the physical hardware, because as previously noted, this has real-time performance implications. The colour buffer may be

 One byte per pixel (indexed or pseudo-colour), which allows 28 = 256 colours, Two bytes per pixel (high colour), which, using 16 bits = 65,536 colours,

**2.3.1 The issues that must be considered in real-time imaging system** 

engineering (Lukac & Plataniotis, 2007).

**2.3.1.1 Hardware and display issues** 

eye to perceive continuous motion.

**2.3.1.2 Colour representation and real-time performance** 

implemented using one or more of the following storage formats:

three primary (RGB) colours at the same time. Each sensor cell has its own spectrally selective filter and thus it stores only a single measurement. Therefore, the CFA image constitutes a mosaic-like grayscale image with only one colour element available in each pixel location. The two missing colours must be determined from the adjacent pixels using a digital processing solution called demosaicking. Such an architecture represents the most cost-effective method currently in use for colour imaging, and for this reason, it is almost universally utilized in consumer-grade digital cameras (Lukac & Plataniotis, 2007).
