Detecting anomalous behaviors such as abnormal tube displacements and arm movements is critical for ensuring ICU patient safety. This project aims to develop an edge computing system that uses cameras to monitor patients in real time, with a focus on identifying high-risk events like tube dislodgments and dangerous arm motions (e.g., attempts to remove medical devices). We will deploy lightweight deep learning models on edge devices to enable efficient, low-latency analysis directly on the device. Additionally, we will collaborate with hospital partners to validate and deploy the system in real ICU settings, ensuring its practical impact on patient care and clinical workflows.