LabVIEW Real-Time running data acquisition functions provides an environment for complete deterministic control. In contrast, an embedded machine vision system is not deterministic. By nature, vision algorithms are not deterministic. Image content, rather than image size, determines the speed of a particular image-processing step. Any data-driven algorithm is always unbounded. For example, performing particle analysis on an image with three particles will occur faster than an equivalent-sized (resolution) image with seven particles. This difference in computation time increases as the variance in the image increases, and as you perform more time-variant processing and analysis functions.
In order to accommodate for this lack of determinism, you should benchmark your worst-case scenario. In this instance, you analyze an image with the maximum number of particles using the image processing steps and then benchmark for speed. This technique determines the upper bound, or response time, of the system.
In the over 200 functions in IMAQ vision, some functions are less variable. These functions, such as threshold, are not entirely data-driven, and are therefore less time-variant. The threshold function operates on the image with significantly less jitter as long as it is the same image size and the same image type. Many image-processing functions fall under this category because they perform a discrete function on each pixel in the image, regardless of image content.