Comparative study: Enhancing crime event classification in video surveillance through ResNet50 and mobileNetV2 analysis
G.D. Deshpande, Tanmay Devare, A.G. Gunjal · 2025
This study conducts a comparative evaluation of ResNet50 and MobileNetV2 for classifying crime events in video surveillance, utilizing the UCF Crime Dataset. By applying thorough preprocessing and deep learning methods, we assess the performance of both models, highlighting a distinct trade-off: MobileNetV2 achieves higher accuracy, whereas ResNet50 is more effective in reducing loss. Our results indicate that MobileNetV2 achieved an accuracy of 86.67% and a loss of 0.3660. This performance was superior to that of ResNet50, which recorded an accuracy of 83.50% and a loss of 0.6344. These insights are critical for choosing the model which is appropriate based on the specific application needs and pave the way for developing hybrid approaches to improve video surveillance systems, thereby enhancing public safety efforts.