Um estudo comparativo da eficiência de algoritmos de subtração de fundo de imagem aplicados a um ambiente simulado de tráfego veicular

Abstract

The main objective of this work is to conduct a research related to Background Subtraction object detection method and analyze its performance when applying MOG, MOG2, GMG, KNN and CNT modeling techniques in a simulated vehicular traffic environment and identify the approach with the most suitable performance for this type of situation. The methodology used consists of developing an algorithm that will implement each modeling technique in a video that presents moving vehicles, where the results obtained by the subtraction will be compared with each pixels from the reference images, which represents the ideal result (Ground Truth). The data collected from the comparisons are used to calculate the accuracy and processing time metrics, which are the indicators used for performance analysis and selection of the modeling technique. The results of the experiment demonstrated that MOG2 technique is the most appropriate for the proposed scenario, with the best precision (66.16%), accuracy (98.59%) and suitable average processing time per frame (8.15 ms) when compared to the other techniques evaluated.