БЕЙНЕ АҒЫНДАРЫНДА ҚОЗҒАЛАТЫН ОБЪЕКТІЛЕРДІ АНЫҚТАУ ҮШІН НАҚТЫ УАҚЫТТАҒЫ МАТЕМАТИКАЛЫҚ МОДЕЛЬДЕУ

Abstract

The development of mathematical models for real-time detection of moving objects in video streams is crucial, particularly in response to the increasing need for security and automation. This study presents a detection system based on the Faster Region-Based Convolutional Neural Network (Faster R-CNN) framework, integrating multiple backbone architectures such as Residual Network-50 (ResNet-50), Feature Pyramid Network (FPN), MobileNet Version 3 Large, and Efficient Network-B0 (EfficientNetB0) enhanced with a Self-Attention mechanism. The model leverages deep neural networks combined with scaling techniques to extract object features effectively, ensuring adaptability across various operational conditions. Experimental findings reveal outstanding accuracy, with the MobileNet Version 3 Large model achieving 95.70% accuracy and the Residual Network-50 model attaining 100% accuracy within the first three epochs. The Faster R-CNN model using EfficientNetB0 with a Self-Attention mechanism also achieves 100% accuracy by the third epoch, maintaining consistent performance in subsequent training cycles. The model with the ResNet-50 and FPN backbone demonstrates a reduction in average loss from 0.0922 during the first epoch to 0.0102 by the 15th epoch, highlighting its robustness and efficiency. Future investigations may focus on enhancing the proposed system to manage more complex and dynamic video scenarios, as well as optimizing data processing to lower computational costs and improve processing speed.