While crane operations in manufacturing are inherently high-risk, safety monitoring remains predominantly manual -even when supported by video surveillance- and this approach has proven insufficient to reduce industrial accident rates. To address this limitation, this study proposes a smart monitoring system using RGB-D data to automatically detect hazardous behaviors: "Distraction" and "Contact". For the Distraction detection, the system integrates YOLO-face, Gaze360, and LiDAR data to quantify 3D attention clusters, prioritizing distracted gaze during load movement to reduce false alarms. For the Contact detection, YOLO-Pose and ByteTrack enable consistent 3D coordinate tracking of operators and loads, approximating the load as a geometric model to identify physical proximity. This novelty lies in achieving robust 3D behavior understanding using a single monocular RGB-D sensor in dynamic environments. Experimental results demonstrate the feasibility of the proposed framework, providing a foundation for future real-time industrial safety monitoring.