Intelligent cockpit driver monitoring systems play a vital role in modern vehicle safety, especially in the context of the gradual popularisation of autonomous driving technology. This paper proposes an intelligent cockpit driver monitoring system based on deep learning. Through multi-task learning, multi-modal data fusion, and dynamic resource scheduling technology, it achieves real-time monitoring of driver fatigue, distraction behaviour, emotional state, and driving behaviour. The system uses a variety of deep learning models, including improved MobileNet, dual-stream CNN, LSTM-GCN, etc. and has demonstrated excellent performance on multiple key tasks. Through optimisation strategies such as knowledge distillation, network pruning, and quantisation, the system maintains efficient operation in the embedded in-vehicle environment. After large-scale simulation testing, the system has outperformed existing solutions on the market in terms of stability, robustness, and response speed. The research in this paper not only provides technical support for improving driving safety, but also lays the foundation for the development of future intelligent cockpit technology.