<p>Suspicious activity detection is crucial for public safety, enabling the prevention of crimes such as theft, robbery, and harassment. The rapid advancement of edge computing has enabled real-time data processing at the source, reducing latency and bandwidth constraints. The core idea of this work is to develop a Smart Surveillance System that combines Edge-Cloud-IoT architecture for automated suspicious activity recognition. Therefore, this paper introduces a novel deep learning model named Circular Queue-Guided Attentive Stacked Parallel Convolution Network to enhance suspicious activity recognition in surveillance videos. The Stacked Parallel Convolution module employs a variety of kernel sizes to extract multi-scale features, while the Co-Attention Mechanism improves feature representation. A Circular Queue is used to reduce inter-class similarity problems in convolutional neural networks and increase classification accuracy by utilizing temporal information. The decision-making procedure improves recognition reliability by depending on the classification labels that appear most frequently during a given time span. Real-world surveillance footage is used for experimental evaluations, which are processed on a Raspberry Pi edge computing architecture with cloud backup provided by Amazon Web Services. The experimental results provided that the proposed model achieved an accuracy of 99.18%, outperforming existing approaches across metrics such as recall, precision, and F1-score. The proposed model offers a practical solution for real-time monitoring, with reduced response time and power consumption compared to traditional methods. The result demonstrates the model’s potential for real-time automated surveillance applications.</p>