This document proposes using edge computing and multiple deep learning models for improved AI surveillance. The models include face detection, landmarks recognition, face re-identification, and Mask R-CNN for object detection. These models would be deployed on edge devices using the Intel OpenVino toolkit to perform real-time surveillance with low latency. Experimental results show the edge computing approach can process video frames at 25 FPS for smart classroom monitoring, compared to 10 FPS for cloud-based approaches. Initial testing of the Mask R-CNN model achieved a validation loss of 0.2294 for weapon detection. The proposed system aims to enhance security monitoring while reducing resources required compared to cloud-based solutions.