Impact of Camera Characteristics and Settings on Precession of AI Object Recognition Models

This paper provides an observation about performance of models of object detection in different conditions of input image quality. Now, there are many different computer vision models suitable for special tasks. Therefore, some of them may not be accurate in some forecasts, but they do not miss a single true value. When choosing a computer vision model, it is necessary to rely on the problems that need to be solved, since there is no universal solution. However, in this article, the choice of models is based on the Average Precision (AP) metric - which allows us to highlight how the model copes with different types of tasks, without losing its accuracy. The MS COCO (Microsoft Common Objects in Context) database was taken as the basis for the measurements, since it is one of the most extensive and rich databases which is used to compare the characteristics of various computer vision technologies, and its dataset covers most areas. The article also discusses metrics and analysis performance, based on what model was selected.