In developed countries, the aging population is increasing steadily, raising concerns related to healthcare and patient safety. One critical issue is the growing number of bed falls and injury incidents in hospitals and nursing homes, often occurring when patients attempt to get out of bed unassisted. Accurately detecting such movements and promptly notifying nursing staff are essential for preventing accidents. This work proposes a camera-based bed fall prevention system that utilizes the YOLOv11 deep learning object detection model and a single infrared camera to detect four key patient postures: supine, side-lying, getting up, and edge-sitting. Previous research has shown that factors such as blanket occlusion and presence of caregiver complicate accurate posture recognition. To address these challenges, we developed a multilabel dataset and trained a custom YOLOv11 model capable of simultaneously detecting patient posture, head position, and bed location. Training data were collected in four different scenarios, and system evaluation was conducted using three datasets: (1) simulated laboratory data from 19 participants, (2) real hospital data from five elderly participants; and (3) hospital data captured under various lighting conditions. The system successfully identifies the “getting-up” posture, tracks head movement beyond bed boundaries, issues alerts, estimates the number of visible heads, and recognizes safety modes. Experimental results demonstrate high performance, with an average accuracy of 99.4%, sensitivity of 98.8%, and specificity of 99.7%. The proposed system offers a practical and cost-effective solution for bed fall prevention, with the potential to reduce the workload of clinical staff and improve patient safety.