For centuries, technological development has been one of the main goals of replacing human workflows with machines to provide more efficient, cheaper and safer products. The assembly processes are replaced by robot arms, the moving by vehicles, the vision by cameras. Human vision is not only the reception of visual information that the cameras are intended to perceive, but the processing of information has to be solved as well. For processing image information, we apply computer vision systems that are able to interpret images captured by the camera and even to make decisions accordingly. Computer vision systems help people in many areas, such as crime prevention, medicine, automotive industry, smart city development, and so on.
In my dissertation first I present the computer vision systems and challenges in general. Additionally, I describe the steps of object tracking, with emphasis on pedestrian tracking. I show step by step how to create the object list from corresponding images starting from the background substraction, through the error reduction to determining the path of moving pedestrians.
Furthermore, I describe the architecture of the developed system with pointing to the component relationships and the purpose of components. I show how I followed the coding principles when I designed the system, such as correct naming, optimized functions, reusability, readability, etc.
I examine the completed application efficiency between the reference measurements that compare the human vision with the capabilities of the application. To this end, a set of reference measurements was made so that it could be compared with the results of the application for the same series of images.
Finally, I demonstrate how to test a computer vision system that places great emphasis on reproducible testing steps in order to efficiently align the parameters needed to operate the application. In addition, I demonstrate some test methods, comparing its advantages and disadvantages.