Performance engineering of image processing systems through benchmarking techniques

Currently, a multitude of systems are developed that process a continuous flow of images. They are used, for example, for video surveillance, inspection and control of products manufactured in production lines, automatic guidance of robots, vehicles, etc. For these systems to be useful, they must be able to process and store images at a minimum frequency. Therefore, when designing these systems it will be essential to determine the frequency at which they can operate depending on the characteristics of the images, the algorithms used to process them, and the computer hardware selected. Despite the importance of estimating and adjusting the performance of these imaging systems, there are hardly any methodologies for developing the performance engineering of them. With the aim of covering this gap, this article presents a simple experimental method to develop the performance engineering of the image processing systems and in particular to determine their maximum operational frequency or throughput. Using the proposed method, any performance engineer can determine the maximum working frequency of a system systematically and check that it exceeds the minimum required frequency. In addition, the engineer can also obtain the necessary information to reconfigure the system, eliminate performance bottlenecks, and take advantage of the computing power of the hardware properly.