Development of a methodology for selecting machine learning models for recognizing moving objects in a video stream based on production rules

To date, technologies in the field of photo-video data processing, algorithms for recognizing and classifying objects in images, have become more accurate, often exceeding the accuracy threshold of 90%. Such technological breakthroughs have led to extensive use of these innovations for professional and personal purposes. However, the variation of factors such as the conditions for obtaining and processing images, the dynamics and deformation of objects in the frame, can complicate the practical application of these algorithms. The presented scientific research focuses on the development of a recommendation system for choosing optimal machine learning models in order to solve a wide range of tasks related to object recognition. The principle of forming recommendations is generated on the basis of production rules, which are developed taking into account experimental data and analysis of academic sources. The result of the proposed system is not just a list of models indicating their relevance, but also proposals for the creation of machine learning pipelines and recommendations for the installation and use of appropriate program libraries. The current article describes a methodology for automating the formation of recommendations, including a priori estimation of metric values in the context of object classification tasks in images.