STRUCTURED DECISION-MAKING METHOD FOR CHOOSING A MODIFICATION OF A ROAD INFRASTRUCTURE OBJECT RECOGNITION MODEL

The article presents the problem of selecting a modification of the road infrastructure object recognition model. To solve this problem, it is proposed to use a structured decision-making method based on multi-criteria assessments. The main attention is paid to identifying and comparative analysis of the most significant criteria for model evaluation, such as Precision, Recall and F1-score metrics. During the study, pairwise comparison matrices were constructed, which allow not only to visually represent the relative importance of each parameter, but also to quantitatively assess their impact on the overall efficiency of the model. The process of forming pairwise comparison matrices includes the opinion of experts in the field of machine learning and computer vision, which ensures a high degree of reliability of the results. The pairwise comparison matrix was assessed on an intensity scale from 1 to 9 (1 - equal, 3 - slightly better, 5 - better, 7 - significantly better, 9 - fundamentally better). For each criterion, the values of normalized estimates of the priority vector, consistency index and consistency ratio were determined. The decision-making condition for calculating criteria was assessed by the value of the consistency ratio. In case of non-compliance with the condition, the matrices of pairwise comparisons were recalculated taking into account the expert's assessment. After performing calculations, including the weighting of each criterion, priorities were derived for selecting a model for recognizing road infrastructure objects. As a result of the calculations, it was revealed that of the five modifications, the YOLOv8n model has the highest priority. The selected model was tested using the developed mobile application "Driver Assistant" on smartphones with the Android operating system for recognizing road infrastructure objects by class (road sign, road marking, traffic lights) in real time.