Tower crane operations represent one of the highest-risk activities in the construction industry, with operator fatigue, cognitive overload, and inadequate hazard perception being the leading contributory factors in fatal and non-fatal accidents globally. This study presents the design, implementation, and empirical evaluation of an AI-Based Tower Crane Operator Safety Monitoring System (AI-TCOSMS) that integrates computer vision, deep learning, Internet of Things (IoT) sensing, and real-time data analytics to continuously monitor operator physiological and behavioural parameters. A mixed-method quasi-experimental research design was employed across six active construction sites over six months, involving 87 crane operators. The system deployed a hybrid CNN-LSTM deep learning model for fatigue detection, achieving 95.7% classification accuracy, significantly outperforming both standalone CNN (89.4%) and LSTM (91.2%) architectures. Post-implementation data revealed a 46.4% reduction in workplace accidents, a 38.6% decrease in crane downtime, and an average operator safety compliance improvement of 34%. Operator acceptance surveys recorded a mean satisfaction score of 4.1 out of 5.0. Statistical analyses, including paired t-tests and ANOVA, confirmed that all five research hypotheses were supported at the p < 0.05 significance level. The findings demonstrate that AI-driven safety monitoring systems can substantially reduce occupation hazards in crane-intensive construction environments, with broader implications for smart construction cite management and regularity framework.