A Robust Object Detection Model for Real-Time Surveillance and Crime Detection in Public Spaces

The rise in violence and criminal activities necessitates advanced surveillance systems capable of rapid and accurate threat detection. This paper proposes a robust object detection system combining the speed of YOLO (You Only Look Once) with the accuracy of Faster R-CNN (Region-based Convolutional Neural Networks) to enhance public safety. The system operates in two stages: YOLO performs fast initial detections, followed by refined detection by Faster R-CNN to minimize false positives and false negatives. The system was evaluated on multiple datasets, including COCO, Pascal VOC, UCF-Crime, and CASIA-CCTV, achieving superior results across key performance metrics. The hybrid model attained an average precision of 99.25%, an F1 score of 99%, and a processing speed of 60 frames per second (FPS), significantly outperforming individual models. Additionally, the system's hybrid storage solution optimized storage capacity, reducing storage requirements by 98%. This research demonstrates that the proposed system offers a significant improvement in both detection accuracy and efficiency, contributing to the development of safer public spaces. Real-time alerts and efficient data management ensure timely responses to security threats, making this system highly practical for modern surveillance applications.