SMART VISION SYSTEMS FOR PUBLIC SAFETY: REAL-TIME CROWD MONITORING AND ANOMALY DETECTION IN URBAN SPACES USING DEEP LEARNING AND EDGE COMPUTING

As the urban populations grow rapidly and public security demands increase, intelligent surveillance systems have emerged as a key part of the smart city infrastructure. Conventional CCTV monitoring is often dependent on human operators, which is inefficient and liable to missed or delayed detections. This paper presents ConvLSTM, an online deep learning model, which is targeted for intelligent crime surveillance and anomaly detection in densely populated urban areas. The suggested system combines several AI modules—YOLOv8 for object detection, ConvLSTM for anomaly detection into an integrated and optimized pipeline. The essential innovation of this system comes from its hybrid architecture supporting Spatio-temporal behavior analysis with real-time processing capability. Deployed on Python 3.10, PyTorch, TensorFlow, OpenCV, and running on Python, the system offers a very efficient edge deployment with local processing as well as the strengthened privacy. Processing video inputs is done via detection, tracking, and anomaly detection analysis steps on urban environments, which support automatic alerts without needing cloud resources. The model is trained and validated on the Kaggle dataset. The model yields a detection accuracy of 97.3%, better than competitive approaches like YOLOv9 and ConvNextv1 by 1.2–2.0%, while also sustaining a real-time frame rate of 20 FPS. These observations validate the model's ability to process dynamic urban environments with high accuracy, recall, and low latency. In conclusion, offers a high-performance, privacy-aware, and scalable solution for crowd monitoring and real-time anomaly detection in contemporary intelligent cities. The proposed Method will be useful on anomaly detection on the urban environments.