Foreground Detection in Dynamic Scenes Using Convolutional Neural Networks in Comparison with Gaussian Mixture Model to Measure F-score

This study focuses on evaluating the performance of Convolutional Neural Networks in detecting moving objects, particularly in scenarios where their accuracy and F-Score may be suboptimal. We use ClinCalc software to calculate sample sizes and analyze results. Our approach involves applying CNNs for foreground identification using the CDNET 2014 dataset. This dataset allows us to compare the performance of the Convolutional Neural Network with that of the Gaussian Mixture Model. Our findings reveal that CNNs can significantly enhance object detection performance. Specifically, CNNs achieved an impressive mean F-Score of 89% and an accuracy rate of 96%. These results underscore the effectiveness of CNNs in improving object detection, especially when compared to traditional methods. By leveraging CNNs to perform background subtraction and identify moving objects, we demonstrate their superior capability in handling complex visual data. This approach not only highlights the strengths of CNNs in object detection tasks but also provides a benchmark for future research and application in this domain.